| Language Model | Executability | LCS | Correctness |
| Vanilla GPT-2117M | 18.66% | 3.19% | 14.27% |
| Vanilla GPT-21.5B | 39.40% | 7.78% | 19.51% |
| Vanilla Codex 2.5B | 17.62% | 15.57% | 53.44% |
| Vanilla GPT-Neo 2.7B | 29.92% | 11.52% | 50.74% |
| Vanilla Codex 12B | 18.07% | 16.97% | 56.06% |
| Vanilla GPT-312B Vanilla GPT-3175B | 25.87% 7.79% | 13.40% | 30.98% |
| Annotated GT | 100.00% | 17.82% N/A | 65.19% 54.80% |
| Fine-tuned GPT-3 12B | 66.07% | 34.08% | 50.56% |
| OurFinal Methods | |||
| Translated Codex 12B | 78.57% | 24.72% | 34.97% |
| Translated GPT-3175B | 73.05% | 24.09% | 51.73% |
| Methods | Executability | LCS |
| Translated Codex 12B | 78.57% | 24.72% |
| - w/o Action Translation | 31.49% | 22.53% |
| - w/o Dynamic Example | 50.86% | 22.84% |
| - w/o Iterative | 55.19% | 24.43% |
| Methods | # ofC | #of C and E | E/C |
| GPT-21.5B | 8 | 0 | 0.00% |
| GPT-312B | 24 | 2 | 8.33% |
| GPT-3175B | 68 | 5 | 7.35% |
| Codex 12B | 45 | 8 | 17.78% |
| Translated Codex 12B | 18 | 15 | 83.33% |
| Name | Description | Search Values |
| epsilon (e) | OOD cutoff threshold used in iterative action{0,O.4,0.8} translation | |
| temperature | sampling parameter adjusting relative probabili-{0.1, O.3, 0.6} ties across tokens | |
| k | number of samples generated when querying LMs{1, 10} each time | |
| frequence_penalty | OpenAI API specific; penalize new tokens based on their existing frequency in the text so far | {0.1, 0.3, 0.6, 0.9} |
| presence_penalty | OpenAI API specific; penalize new tokens based on whether they appear in the text so far | {0.3,0.5, 0.8} |
| repetition_penalty | Hugging Face Transformers specific； penalize new tokens based on whether they are repeating existing text | {1.0, 1.2, 1.5, 1.8} |
| Example 1 | Example 2 | Example 3 | |
| Task: Use computerStep 1: Walk to home officeStep 2: Walk to chairStep 3: Find chairStep 4: Sit on chairStep 5: Find computerStep 6: Switch on computerStep 7: Turn to computerStep 8: Look at computerStep 9: Find keyboardStep 10: Type on keyboard | Task: Relax on sofaStep 1: Walk to home officeStep 2: Walk to couchStep 3: Find couchStep 4: Sit on couchStep 5: Find pillowStep 6: Lie on couch | Task:Read bookStep 1: Walk to home officeStep 2: Walk to novelStep 3: Find novelStep 4: Grab novelStep 5: Find chairStep 6: Sit on chairStep 7: Read novel | |
| Methods | Executability | Average Length |
| Vanilla GPT-2 1.5B | 39.40% | 4.24 |
| Vanilla Codex 12B | 18.07% | 7.22 |
| Translated Codex 12B | 78.57% | 7.13 |
| Ground-Truth | 100.00% | 9.66 |
| Annotated Ground-Truth | Codex | Translated Codex (Ours) |
| Task: Go to sleep Step 1: Walk to bedroom Step 2: Walk to bed Step 3: Find bed Step 4: Turn to bed | Task: Go to sleep Step 1: Walk to bedroom Step 2: Walk to bed Step 3: Find bed Step 4: Sit on bed | Task: Go to sleep Step 1: Walk to bedroom Step 2: Walk to bed Step 3: Lie on bed Step 4: Sleep |
| Task:Browse internet Step 1: Walk to home office Step 2: Walk to chair Step 3: Find chair Step 4: Sit on chair Step 5: Find computer Step 6: Switch on computer Step 7: Turn to computer Step 8: Look at computer Step 9: Find keyboard Step 10: Type on keyboard Task:Listen to music | Task:Browse internet Step 1: Walk to home office Step 2: Walk to chair Step 3: Find chair Step 4: Sit on chair Step 5: Find computer Step 6: Switch on computer Step 7: Turn to computer Step 8: Look at computer Step 9: Find keyboard | Task:Browse internet Step 1: Walk to home office Step 2: Walk to desk Step 3: Find chair Step 4: Sit on chair Step 5: Find computer Step 6: Switch on computer Step 7: Find mouse Step 8: Grab mouse Step 9: Turn to computer Step 10: Point at computer |
| Step 1: Walk to bedroom Step 2: Walk to computer Step 3: Find computer Step 4: Turn to computer Step 5: Look at computer Step 6: Walk to computer Step 7: Find chair Step 8: Pull chair Step 9: Sit on chair Step 10: Find headset Step 11: Grab headset Step 12: Put on headset Step 13: Switch on computer Step 14: Find keyboard Step 15: Type on keyboard Task:Brush teeth Step 1: Walk to bathroom Step 2: Walk to tooth paste Step 3: Find tooth paste Step 4: Grab tooth paste Step 5: Find toothbrush Step 6: Grab toothbrush Step 6:Move brush around mouth for two minutes Step 7: Spit out toothpaste and brush into sink | Step 10: Type on keyboard Task:Listen to music Step 1: Walk to home office Step 2: Walk to chair Step 3: Find chair Step 4: Sit on chair Step 5: Find computer Step 6: Switch on computer Step 7: Turn to computer Step 8: Look at computer Step 9: Find keyboard Step 10: Type on keyboard Task:Brush teeth Step 1: Walk to bathroom Step 2: Walk to sink Step 3: Find toothbrush Step 4: Pick up toothbrush Step 5: Put toothbrush in mouth | Task:Listen to music Step 1: Walk to home office Step 2: Walk to computer Step 3: Sit on chair Step 4: Switch on computer Step 5: Find music stand Task:Brush teeth Step 1: Walk to bathroom Step 2: Open door Step 3: Walk to sink Step 4: Put pot on sink Step 5: Put brush on toothbrush |
| Task: Dry soap bottles Step 1: Find towel Step 2: Find soap Step 3: Grab soap Step 4: Wipe soap Step 5: Put back soap Step 6: Find shampoo Step 7: Grab shampoo Step 8: Wipe shampoo Step 9: Put back shampoo Step 10: Find shampoo Step 11: Grab shampoo Step 12: Wipe shampoo Step 13: Put back shampoo | Step 8: Turn on water in sink and rinse brush for one minute Step 9: Turn off water in sink and return brush to cupboard Task:Dry soap bottles Step 1: Walk to kitchen Step 2: Walk to sink Step 3: Find sink Step 4: Turn on water tap Step 5: Wait for water to be hot enough Step 6: Fill container with hot water Step 7: Find soap bottles Step 8: Pour soap into container with hot water Step 9: Wait for soap to dissolve in water | Task: Dry soap bottles Step 1: Walk to bathroom Step 2: Walk to shower Step 3: Find soap |
| Task: Vacuum Task: Vacuum Task:VacuumStep 1: Walk to home office Step 1: Walk to hallway Step 1: Walk to home officeStep 2: Walk to computer Step 2: Walk to closet Step 2: Walk to dresserStep 3: Find vacuum cleaner Step 3:Find closet Step 3: Open dresserStep 4:Switch on vacuum Step 4: Open closet Step 4: Find vacuum cleanercleaner Step 5:Find vacuum cleaner Step 5: Grab vacuum cleaner Step 5:Switch off vacuum Step 6: Turn on vacuum cleaner Step 6: Pull vacuum cleanercleaner Step 7: Walk to living room Step 7: Plug in vacuum cleanerStep 8: Vacuum floor in living Step 8: Switch on vacuumroom cleanerStep 9: Pull vacuum cleanerStep 10: Push vacuum cleaner | |||
| Task: Vacuum Step 1: Walk to hallwayStep 2: Walk to closetStep 3:Find closetStep 4: Open closetStep 5:Find vacuum cleanerStep 6: Turn on vacuum cleaner Step 7: Walk to living roomStep 8: Vacuum floor in livingroom | |||
| Task: Find dictionaryStep 1: Walk to home officeStep 2: Walk to bookshelf | Task: Find dictionaryStep 1: Walk to home officeStep 2: Walk to chairStep 3: Find chairStep 4: Sit on chairStep 5: Find dictionary Step 6: Open dictionaryStep 7: Look at dictionary | Task: Find dictionaryStep 1: Walk to home officeStep 2: Walk to bookStep 3: Find textbook | |
| f | |||
| Step 3: Turn to bookshelfStep 4: Look at bookshelfStep 5: Turn to bookshelfStep 6: Look at bookshelfStep 7: Turn to bookshelfStep 8: Look at bookshelfStep 9: Turn to bookshelfStep 10: Look at bookshelfStep 11: Turn to bookshelfStep 12: Look at bookshelfStep 13: Turn to bookshelfStep 14: Look at bookshelfStep 15: Find novelStep 16: Grab novel | |||
| Task:Wash monitorStep 1: Walk to home officeStep 2: Walk to computerStep 3: Find computerStep 4: Turn to computerStep 5: Look at computerStep 6: Switch off computerStep 7: Find brushStep 8: Turn to brushStep 9: Look at brushStep 10: Grab brush Step 11: Wipe computerStep 12: Drop brushStep 13:Find napkinStep 14: Grab napkinStep 15: Wash napkinStep 16: Wipe computer | Task:Wash monitorStep 1: Walk to home officeStep 2: Walk to chairStep 3:Find chairStep 4: Sit on chairStep 5: Find computerStep 6: Switch on computerStep 7: Turn to computerStep 8: Look at computerStep 9: Find keyboardStep 10: Type on keyboard | Task:Wash monitorStep 1:Walk to checkStep 2: Walk to desk | |
| Task: Read newspaperStep 1: Walk to home officeStep 2: Walk to chairStep 3: Find chairStep 4: Sit on chairStep 5: Find newspaperStep 6: Grab newspaperStep 7: Open newspaperStep 8: Read newspaper | Task: Read newspaperStep 1: Walk to home officeStep 2: Walk to chairStep 3: Find chairStep 4: Sit on chairStep 5: Find newspaperStep 6: Open newspaperStep 7: Read newspaper | Task: Read newspaperStep 1: Walk to home officeStep 2: Walk to chairStep 3: Find chairStep 4: Sit on chairStep 5: Find newspaperStep 6: Grab newspaperStep 7: Read newspaper | |
| Task:ApplylotionStep 1: Find bedStep 2: Turn to bedStep 3: Point at bedStep 4: Wash bedStep 5: Turn to bedStep 6: Look at bedStep 7: Push bedStep 8: Sit on bed | Task:ApplylotionStep 1: Walk to bathroomStep 2: Walk to cabinetStep 3:Open cabinetStep 4: Find lotionStep 5: Apply lotion on handsStep 6: Rub hands together for10 secondsStep 7: Apply lotion on face | Task:ApplylotionStep 1: Walk to bathroomStep 2: Turn to lotionbottleStep 3:Grab lotionbottle | |
| Task: Draw pictureStep 1: Walk to home officeStep 2: Walk to chairStep 3: Find chairStep 4: Sit on chairStep 5: Find deskStep 6: Open deskStep 7: Find checkStep 8: Grab checkStep 9: Put check on deskStep 10: Find pencilStep 11: Grab pencilStep 12: Close desk | Task: Draw pictureStep 1: Walk to home officeStep 2: Walk to chairStep 3: Find chairStep 4: Sit on chairStep 5: Find computerStep 6: Switch on computerStep 7: Turn to computerStep 8: Look at computerStep 9: Find keyboardStep 10: Type on keyboard | Task: Draw pictureStep 1: Walk to home officeStep 2: Walk to wallStep 3: Walk to drawingStep 4: Find drawingStep 5: Turn to drawingStep 6: Look at drawingStep 7: Touch drawingStep 8: Pull drawingStep 9: Push drawing | |
| Task: Prepare pot of boiling wa-terStep 1: Walk to dining roomStep 2: Walk to potStep 3: Find potStep 4: Grab potStep 5: Walk to sinkStep 6: Find faucetStep 7: Switch on faucetStep 8: Put pot on sinkStep 9: Grab potStep 10: Switch off faucetStep 11: Grab potStep 12: Walk to oven Step 13: Put pot on ovenStep 14: Switch on oven | Task: Prepare pot of boiling wa-terStep1: Walk to kitchenStep 2: Walk to sinkStep 3:Find sink Step 4: Turn on tapStep 5: Fill kettle with waterStep 6: Put kettle on stoveStep 7: Turn on stoveStep 8: Wait for water to boil | Task: Prepare pot of boiling wa-terStep 1: Walk to kitchenStep 2: Walk to cabinetStep 3: Find pot Step 4: Grab pot Step 5: Walk to stoveStep 6: Find stoveStep 7: Switch on stove | |
| Task: Look at mirrorStep 1: Walk to bathroomStep 2: Walk to mirrorStep 3:Find mirrorStep 4: Turn to mirrorStep 5: Look at mirror | Task: Look at mirrorStep 1: Walk to bathroomStep 2: Walk to mirrorStep 3: Look at mirror | Task:Look at mirrorStep 1: Walk to bedroomStep 2: Walk to mirrorStep 3:Find mirrorStep 4: Turn to mirrorStep 5: Look at mirror | |
| Task:Wipe down sinkStep 1: Walk to dining roomStep 2: Walk to ragStep 3: Find ragStep 4: Grab ragStep 5: Walk to sinkStep 6: Wipe sinkStep 7: Put back rag | Task:Wipe down sinkStep1: Walk to kitchenStep 2: Walk to sinkStep 3: Find sinkStep 4: Wipe down sinkStep 5: Return to kitchen | Task:Wipe down sinkStep1: Walk to bathroomStep 2: Walk to cleaning solutionStep 3: Find cleaning solutionStep 4: Grab cleaning solutionStep 5: Find sinkStep 6: Pour cleaning solutioninto glassesStep 7: Find faucetStep 8: Switch on faucetStep 9: Scrub sinkStep 10:Wash sink | |
| Task: Complete surveys on ama- Task: Complete surveys on ama- Task: Complete surveys on ama-zon turk zon turk zon turkStep 1: Walk to chair Step 1: Walk to home office Step 1: Walk to home officeStep 2: Sit on chair Step 2: Walk to chair Step 2: Walk to desk Step 3: Find computer Step 3: Find chair Step 3: Find chairStep 4: Switch on computer Step 4: Sit on chair Step 4: Sit on chairStep 5: Turn to computer Step 5: Find computer Step 5: Find computerStep 6: Look at computer Step 6: Switch on computer Step 6: Switch on computerStep 7: Find mouse Step 7: Turn to computerStep 8: Grab mouse Step 8: Look at computerStep 9: Turn to computer Step 9:Find keyboard Step 10: Point at computer Step 10: Type on keyboardStep 11: Find checkStep 12: Grab checkStep 13: Read checkStep 14: Put back mouseStep 15: Find keyboardStep 16: Type on keyboard | |||
| Task: Complete surveys on ama-zon turkStep 1: Walk to chair | Task: Complete surveys on ama- zon turkStep 1: Walk to home officeStep 2: Walk to chairStep 3: Find chairStep 4: Sit on chairStep 5: Find computerStep 6: Switch on computerStep 7: Turn to computerStep 8: Look at computerStep 9:Find keyboardStep 10: Type on keyboard | ||
| Step 2: Sit on chair Step 3: Find computerStep 4: Switch on computerStep 5: Turn to computer | |||
| Step 6: Look at computerStep 7: Find mouseStep 8: Grab mouseStep 9: Turn to computer Step 10: Point at computerStep 11: Find checkStep 12: Grab checkStep 13: Read checkStep 14: Put back mouseStep 15: Find keyboardStep 16: Type on keyboard | |||
| Task:Write book Step 1: Walk to home office Step 2: Walk to laptopStep 3: Find laptopStep 4: Grab laptopStep 5: Find electrical outletStep 6: Plug in laptopStep 7: Walk to love seatStep 8: Sit on love seat | Task: Write bookStep 1: Walk to home officeStep 2: Walk to chairStep 3: Find chairStep 4: Sit on chairStep 5: Find computerStep 6: Switch on computerStep 7: Turn to computerStep 8: Look at computerStep 9: Find keyboardStep 10: Type on keyboard | Task:Write bookStep 1: Walk to home officeStep 2: Walk to lightStep 3: Find lightStep 4: Switch on lightStep 5: Find penStep 6: Grab penStep 7: Find chairStep 8: Sit on chairStep 9: Turn to novel | |
| Step 7: Walk to love seatStep 8: Sit on love seatStep 9: Switch on laptopStep 10: Find keyboardStep 11: Type on keyboard | |||
| Task: Paint ceiling Step 1: Find drawing Step 2: Turn to drawingStep 3: Point at drawing Step 4: Watch drawingStep 5: Touch drawingStep 6: Walk to bedroom | Task: Paint ceilingStep 1: Walk to home officeStep 2: Walk to chairStep 3: Find chairStep 4: Sit on chairStep 5: Find paint canStep 6: Open paint can Step 7: Pour paint into paint canStep 8: Close paint canStep 9: Find roller Step 10: Roll paint onto ceiling | Task: Paint ceilingStep 1: Walk to home office Step 2: Walk to ceilingStep 3: Walk to paintingStep 4: Find paintingStep 5: Turn to paintingStep 6: Look at paintingStep 7: Touch paintingStep 8: Pull paintingStep 9: Push painting | |
| Task: Change clothesStep 1: Walk to bedroomStep 2: Walk to dresserStep 3: Open dresserStep 4: Find dressStep 5: Grab dressStep 6: Put on dressStep 7: Close dresser | Task: Change clothesStep 1: Walk to closetStep 2: Open closetStep 3: Find clothesStep 4: Put on clothesStep 5: Close closet | Task: Change clothesStep 1: Walk to bedroomStep 2: Walk to dresserStep 3: Find dresserStep 4: Open dresserStep 5: Find pantsStep 6: Grab pantsStep 7: Put on pantsStep 8: Find shirtStep 9: Grab shirtStep 10: Put on shirtStep 11: Close dresser | |
| Task: Make bed Task: Make bed Task:Make bedStep 1: Walk to bedroom Step 1: Walk to bedroom Step 1: Stand upStep 2: Walk to bed Step 2: Walk to bed Step 2: Walk to bedroomStep 3: Find sheets Step 3: Find bed Step 3: Walk to bedStep 4: Grab sheets Step 4: Sit on bed Step 4: Turn to bedStep 5: Pull sheets Step 5: Find sheet Step 5: Put sheets on bedStep 6: Pick up sheetStep 7: Find pillowStep 8: Pick up pillowStep 9: Find blanketStep 10: Pick up blanket | |||
| Step 2: Walk to bedStep 3: Find sheetsStep 4: Grab sheetsStep 5: Pull sheets | |||
| Task: Read yourself to sleepStep 1: Walk to bedroomStep 2: Walk to novelStep 3: Find novelStep 4: Grab novelStep 5: Find bedStep 6: Sit on bedStep 7: Find bookmarkStep 8: Grab bookmarkStep 9: Read novelStep 10: Put bookmark on novelStep 11: Sleep | Task: Read yourself to sleep Step 1: Walk to home officeStep 2: Walk to chairStep 3: Find chairStep 4: Sit on chairStep 5: Find computerStep 6: Switch on computerStep 7: Turn to computerStep 8: Look at computerStep 9: Find keyboardStep 10: Type on keyboard | Task: Read yourself to sleepStep 1: Walk to bedroomStep 2: Walk to bookStep 3: Sit on chairStep 4: Read checkStep 5: Sleep | |
| Task: Hang keysStep 1: Walk to bedroomStep 2: Walk to dresserStep 3: Find purseStep 4: Grab purseStep 5: Open purseStep 6: Find keysStep 7: Grab keysStep 8: Find bowlStep 9: Put keys on bowlStep 10: Close purseStep 11: Put back purse | Task: Hang keysStep 1: Walk to home officeStep 2: Walk to chairStep 3: Find chairStep 4: Sit on chairStep 5: Find computerStep 6: Switch on computerStep 7: Turn to computerStep 8: Look at computerStep 9: Find keyboardStep 10: Type on keyboard | Task: Hang keysStep 1: Walk to home officeStep 2: Walk to tableStep 3: Find keysStep 4: Grab keysStep 5: Walk to wallStep 6: Find hangerStep 7: Put keys on hanger | |
| Task:Eat cheeseStep 1: Walk to dining roomStep 2: Walk to freezerStep 3: Open freezerStep 4: Find cheeseStep 5: Grab cheeseStep 6: Eat cheese | Task:Eat cheeseStep 1: Walk to fridge Step 2: Open fridgeStep 3: Find cheeseStep 4: Pick up cheeseStep 5: Walk to tableStep 6: Place cheese on tableStep 7: Close fridgeStep 8: Walk to chairStep 9: Sit on chairStep 10: Eat cheese | Task: Eat cheeseStep 1: Walk to dining roomStep 2: Walk to tableStep 3: Find tableStep 4: Turn to tableStep 5: Find chairStep 6: Sit on chairStep 7: Find foodStep 8: Grab foodStep 9: Find plateStep 10: Put food on plate | |
| MLLMs | VPG | VPG Trainable | LLM | LLMTrainable |
| KOSMOS-1[19] | CLIP [43] | All | Rand.Init.LM | All |
| Frozen [54] | NF-ResNet-50 [6] | NF-ResNet-50 | GPT-2-like† [42] | No |
| Flamingo [2] | NFNet-F6 [6]+Resampler [21] | Resampler | Chinchilla [18] | Xattn-Dense |
| PaLM-E[16] | ViT[15]/OSRT[49] | All | PaLM[11] | No |
| BLIP-2 [30] | EVA-CLIP [52] + Q-Former [30] | Q-Former | OPT[60]／Flan-T5[12] | No |
| Epoch | w/ init. | w/o init. |
| 1 | 130.2 | 126.1 |
| 2 | 132.7 | 131.6 |
| 3 | 133.4 | 132.8 |
| Transfer | COCO Caption | NoCaps | VQAv2 | GQA | OKVQA |
| OPT125M→350M | 1.7 | 3.0 | 1.0 | 5.0 | 5.0 |
| OPT125M→1.3B | 9.0 | 10.0 | 9.0 | = | 2.0 |
| OPT350M→1.3B | 4.5 | 5.0 | 9.0 | 2.0 | 2.0 |
| OPT125M→2.7B | 10.0 | 10.0 | 2.0 | 2.0 | 3.0 |
| OPT350M→2.7B | 10.0 | 10.0 | 2.0 | 1 | 3.0 |
| OPT1.3B-→2.7B | 3.3 | 3.3 | 2.0 | - | 1.5 |
| FlanT5base→XL | 1.0 | 1.1 | 3.0 | 4.0 | 2.0 |
| FlanT5xL →OPT2.7B | 5.0 | 5.0 | 2.0 | 2.0 | 3.0 |
| OPT2.7B→FlanT5xL | 1.7 | 2.0 | 1 | 2.0 | 1 |
| Models | VQAv2 val | GQA test-dev | OKVQA test | GPU hours | training data |
| BLIP-2 ViT-G OPT6.7B | 54.3 | 36.4 | 36.4 | 631.5 | 129M |
| BLIP-2 ViT-G OPT2.7B-→6.7B (0urs) | 57.2 | 36.2 | 39.8 | 59.0 | 13.8M |
| VL-LLaMA7B (ours) | 58.1 | 37.5 | 37.4 | 67.1 | 13.8M |
| BLIP-2 ViT-G FlanT5xXL | 65.2 | 44.7 | 45.9 | 684.0 | 121.6M |
| BLIP-2 ViT-G FlanT5xL-→XXL (ours) | 65.2 | 45.0 | 45.0 | 32.4 | 5.3M |
| Rank | Model Elo Rating |
| 1 | LLaMA-Adapter v2 [17] 1023.0 |
| 2 LLaVA [35] | 1019.9 |
| 3 | VL-Vicuna (ours) 1012.1 |
| 4 | MiniGPT-4 [62] 1011.9 |
| 5 | InstructBLIP[13] 999.5 |
| 6 | mPLUG-Owl[59] 996.3 |
| 7 Otter [26] | 981.5 |
| 8 BLIP-2 [30] | 955.8 |
| Transfer | COCO Caption | NoCaps | VQAv2 | GQA | OKVQA |
| OPT125M→350M | 1.1 | 2.7 | 10.0 | 6.0 | 6.0 |
| OPT125M→1.3B | - | 2.7 | 1 | - | 1 |
| OPT350M→1.3B | 1 | 1.7 | 1.0 | 2.0 | 1.0 |
| OPT125M→2.7B | 3.0 | 3.0 | 3.0 | 1.0 | 3.0 |
| OPT350M→2.7B | 3.3 | 4.5 | 1.0 | 1.0 | 1.0 |
| OPT1.3B→2.7B | 2.3 | 3.0 | 1 | 1 | 1 |
| FlanT5base-→XL | 1 | 1.0 | 1.8 | 9.0 | 0.4 |
| Transfer | COCO Caption | VQAv2 | ||
| linear | best | linear | best | |
| FlanT5base → OPT350M | 110.8 | 136.5 | 40.4 | 44.2 |
| FlanT5xL →OPT2.7B | 132.1 | 139.3 | 50.4 | 50.3 |
| OPT350M→FlanT5base | 1.1 | 122.1 | 34.0 | 49.9 |
| OPT2.7B→FlanT5xL | 106.5 | 133.2 | 51.3 | 53.5 |
| MLLMs | VPG | VPG Trainable | LLM | LLMTrainable |
| KOSMOS-1[19] | CLIP [43] | All | Rand.Init.LM | All |
| Frozen [54] | NF-ResNet-50 [6] | NF-ResNet-50 | GPT-2-like† [42] | No |
| Flamingo [2] | NFNet-F6 [6]+Resampler [21] | Resampler | Chinchilla [18] | Xattn-Dense |
| PaLM-E[16] | ViT[15]/OSRT[49] | All | PaLM[11] | No |
| BLIP-2 [30] | EVA-CLIP [52] + Q-Former [30] | Q-Former | OPT[60]／Flan-T5[12] | No |
| Epoch | w/ init. | w/o init. |
| 1 | 130.2 | 126.1 |
| 2 | 132.7 | 131.6 |
| 3 | 133.4 | 132.8 |
| Transfer | COCO Caption | NoCaps | VQAv2 | GQA | OKVQA |
| OPT125M→350M | 1.7 | 3.0 | 1.0 | 5.0 | 5.0 |
| OPT125M→1.3B | 9.0 | 10.0 | 9.0 | = | 2.0 |
| OPT350M→1.3B | 4.5 | 5.0 | 9.0 | 2.0 | 2.0 |
| OPT125M→2.7B | 10.0 | 10.0 | 2.0 | 2.0 | 3.0 |
| OPT350M→2.7B | 10.0 | 10.0 | 2.0 | 1 | 3.0 |
| OPT1.3B-→2.7B | 3.3 | 3.3 | 2.0 | - | 1.5 |
| FlanT5base→XL | 1.0 | 1.1 | 3.0 | 4.0 | 2.0 |
| FlanT5xL →OPT2.7B | 5.0 | 5.0 | 2.0 | 2.0 | 3.0 |
| OPT2.7B→FlanT5xL | 1.7 | 2.0 | 1 | 2.0 | 1 |
| Models | VQAv2 val | GQA test-dev | OKVQA test | GPU hours | training data |
| BLIP-2 ViT-G OPT6.7B | 54.3 | 36.4 | 36.4 | 631.5 | 129M |
| BLIP-2 ViT-G OPT2.7B-→6.7B (0urs) | 57.2 | 36.2 | 39.8 | 59.0 | 13.8M |
| VL-LLaMA7B (ours) | 58.1 | 37.5 | 37.4 | 67.1 | 13.8M |
| BLIP-2 ViT-G FlanT5xXL | 65.2 | 44.7 | 45.9 | 684.0 | 121.6M |
| BLIP-2 ViT-G FlanT5xL-→XXL (ours) | 65.2 | 45.0 | 45.0 | 32.4 | 5.3M |
| Rank | Model Elo Rating |
| 1 | LLaMA-Adapter v2 [17] 1023.0 |
| 2 LLaVA [35] | 1019.9 |
| 3 | VL-Vicuna (ours) 1012.1 |
| 4 | MiniGPT-4 [62] 1011.9 |
| 5 | InstructBLIP[13] 999.5 |
| 6 | mPLUG-Owl[59] 996.3 |
| 7 Otter [26] | 981.5 |
| 8 BLIP-2 [30] | 955.8 |
| Transfer | COCO Caption | NoCaps | VQAv2 | GQA | OKVQA |
| OPT125M→350M | 1.1 | 2.7 | 10.0 | 6.0 | 6.0 |
| OPT125M→1.3B | - | 2.7 | 1 | - | 1 |
| OPT350M→1.3B | 1 | 1.7 | 1.0 | 2.0 | 1.0 |
| OPT125M→2.7B | 3.0 | 3.0 | 3.0 | 1.0 | 3.0 |
| OPT350M→2.7B | 3.3 | 4.5 | 1.0 | 1.0 | 1.0 |
| OPT1.3B→2.7B | 2.3 | 3.0 | 1 | 1 | 1 |
| FlanT5base-→XL | 1 | 1.0 | 1.8 | 9.0 | 0.4 |
| Transfer | COCO Caption | VQAv2 | ||
| linear | best | linear | best | |
| FlanT5base → OPT350M | 110.8 | 136.5 | 40.4 | 44.2 |
| FlanT5xL →OPT2.7B | 132.1 | 139.3 | 50.4 | 50.3 |
| OPT350M→FlanT5base | 1.1 | 122.1 | 34.0 | 49.9 |
| OPT2.7B→FlanT5xL | 106.5 | 133.2 | 51.3 | 53.5 |
| MLLMs | VPG | VPG Trainable | LLM | LLMTrainable |
| KOSMOS-1[19] | CLIP [43] | All | Rand.Init.LM | All |
| Frozen [54] | NF-ResNet-50 [6] | NF-ResNet-50 | GPT-2-like† [42] | No |
| Flamingo [2] | NFNet-F6 [6]+Resampler [21] | Resampler | Chinchilla [18] | Xattn-Dense |
| PaLM-E[16] | ViT[15]/OSRT[49] | All | PaLM[11] | No |
| BLIP-2 [30] | EVA-CLIP [52] + Q-Former [30] | Q-Former | OPT[60]／Flan-T5[12] | No |
| MLLMs | VPG | VPG Trainable | LLM | LLMTrainable |
| KOSMOS-1[19] | CLIP [43] | All | Rand.Init.LM | All |
| Frozen [54] | NF-ResNet-50 [6] | NF-ResNet-50 | GPT-2-like† [42] | No |
| Flamingo [2] | NFNet-F6 [6]+Resampler [21] | Resampler | Chinchilla [18] | Xattn-Dense |
| PaLM-E[16] | ViT[15]/OSRT[49] | All | PaLM[11] | No |
| BLIP-2 [30] | EVA-CLIP [52] + Q-Former [30] | Q-Former | OPT[60]／Flan-T5[12] | No |
| Epoch | w/ init. | w/o init. |
| 1 | 130.2 | 126.1 |
| 2 | 132.7 | 131.6 |
| 3 | 133.4 | 132.8 |
| Epoch | w/ init. | w/o init. |
| 1 | 130.2 | 126.1 |
| 2 | 132.7 | 131.6 |
| 3 | 133.4 | 132.8 |
| Transfer | COCO Caption | NoCaps | VQAv2 | GQA | OKVQA |
| OPT125M→350M | 1.7 | 3.0 | 1.0 | 5.0 | 5.0 |
| OPT125M→1.3B | 9.0 | 10.0 | 9.0 | = | 2.0 |
| OPT350M→1.3B | 4.5 | 5.0 | 9.0 | 2.0 | 2.0 |
| OPT125M→2.7B | 10.0 | 10.0 | 2.0 | 2.0 | 3.0 |
| OPT350M→2.7B | 10.0 | 10.0 | 2.0 | 1 | 3.0 |
| OPT1.3B-→2.7B | 3.3 | 3.3 | 2.0 | - | 1.5 |
| FlanT5base→XL | 1.0 | 1.1 | 3.0 | 4.0 | 2.0 |
| FlanT5xL →OPT2.7B | 5.0 | 5.0 | 2.0 | 2.0 | 3.0 |
| OPT2.7B→FlanT5xL | 1.7 | 2.0 | 1 | 2.0 | 1 |
| Transfer | COCO Caption | NoCaps | VQAv2 | GQA | OKVQA |
| OPT125M→350M | 1.7 | 3.0 | 1.0 | 5.0 | 5.0 |
| OPT125M→1.3B | 9.0 | 10.0 | 9.0 | = | 2.0 |
| OPT350M→1.3B | 4.5 | 5.0 | 9.0 | 2.0 | 2.0 |
| OPT125M→2.7B | 10.0 | 10.0 | 2.0 | 2.0 | 3.0 |
| OPT350M→2.7B | 10.0 | 10.0 | 2.0 | 1 | 3.0 |
| OPT1.3B-→2.7B | 3.3 | 3.3 | 2.0 | - | 1.5 |
| FlanT5base→XL | 1.0 | 1.1 | 3.0 | 4.0 | 2.0 |
| FlanT5xL →OPT2.7B | 5.0 | 5.0 | 2.0 | 2.0 | 3.0 |
| OPT2.7B→FlanT5xL | 1.7 | 2.0 | 1 | 2.0 | 1 |
| Models | VQAv2 val | GQA test-dev | OKVQA test | GPU hours | training data |
| BLIP-2 ViT-G OPT6.7B | 54.3 | 36.4 | 36.4 | 631.5 | 129M |
| BLIP-2 ViT-G OPT2.7B-→6.7B (0urs) | 57.2 | 36.2 | 39.8 | 59.0 | 13.8M |
| VL-LLaMA7B (ours) | 58.1 | 37.5 | 37.4 | 67.1 | 13.8M |
| BLIP-2 ViT-G FlanT5xXL | 65.2 | 44.7 | 45.9 | 684.0 | 121.6M |
| BLIP-2 ViT-G FlanT5xL-→XXL (ours) | 65.2 | 45.0 | 45.0 | 32.4 | 5.3M |
| Models | VQAv2 val | GQA test-dev | OKVQA test | GPU hours | training data |
| BLIP-2 ViT-G OPT6.7B | 54.3 | 36.4 | 36.4 | 631.5 | 129M |
| BLIP-2 ViT-G OPT2.7B-→6.7B (0urs) | 57.2 | 36.2 | 39.8 | 59.0 | 13.8M |
| VL-LLaMA7B (ours) | 58.1 | 37.5 | 37.4 | 67.1 | 13.8M |
| BLIP-2 ViT-G FlanT5xXL | 65.2 | 44.7 | 45.9 | 684.0 | 121.6M |
| BLIP-2 ViT-G FlanT5xL-→XXL (ours) | 65.2 | 45.0 | 45.0 | 32.4 | 5.3M |
| Rank | Model Elo Rating |
| 1 | LLaMA-Adapter v2 [17] 1023.0 |
| 2 LLaVA [35] | 1019.9 |
| 3 | VL-Vicuna (ours) 1012.1 |
| 4 | MiniGPT-4 [62] 1011.9 |
| 5 | InstructBLIP[13] 999.5 |
| 6 | mPLUG-Owl[59] 996.3 |
| 7 Otter [26] | 981.5 |
| 8 BLIP-2 [30] | 955.8 |
| Rank | Model Elo Rating |
| 1 | LLaMA-Adapter v2 [17] 1023.0 |
| 2 LLaVA [35] | 1019.9 |
| 3 | VL-Vicuna (ours) 1012.1 |
| 4 | MiniGPT-4 [62] 1011.9 |
| 5 | InstructBLIP[13] 999.5 |
| 6 | mPLUG-Owl[59] 996.3 |
| 7 Otter [26] | 981.5 |
| 8 BLIP-2 [30] | 955.8 |
| Transfer | COCO Caption | NoCaps | VQAv2 | GQA | OKVQA |
| OPT125M→350M | 1.1 | 2.7 | 10.0 | 6.0 | 6.0 |
| OPT125M→1.3B | - | 2.7 | 1 | - | 1 |
| OPT350M→1.3B | 1 | 1.7 | 1.0 | 2.0 | 1.0 |
| OPT125M→2.7B | 3.0 | 3.0 | 3.0 | 1.0 | 3.0 |
| OPT350M→2.7B | 3.3 | 4.5 | 1.0 | 1.0 | 1.0 |
| OPT1.3B→2.7B | 2.3 | 3.0 | 1 | 1 | 1 |
| FlanT5base-→XL | 1 | 1.0 | 1.8 | 9.0 | 0.4 |
| Transfer | COCO Caption | NoCaps | VQAv2 | GQA | OKVQA |
| OPT125M→350M | 1.1 | 2.7 | 10.0 | 6.0 | 6.0 |
| OPT125M→1.3B | - | 2.7 | 1 | - | 1 |
| OPT350M→1.3B | 1 | 1.7 | 1.0 | 2.0 | 1.0 |
| OPT125M→2.7B | 3.0 | 3.0 | 3.0 | 1.0 | 3.0 |
| OPT350M→2.7B | 3.3 | 4.5 | 1.0 | 1.0 | 1.0 |
| OPT1.3B→2.7B | 2.3 | 3.0 | 1 | 1 | 1 |
| FlanT5base-→XL | 1 | 1.0 | 1.8 | 9.0 | 0.4 |
| Transfer | COCO Caption | VQAv2 | ||
| linear | best | linear | best | |
| FlanT5base → OPT350M | 110.8 | 136.5 | 40.4 | 44.2 |
| FlanT5xL →OPT2.7B | 132.1 | 139.3 | 50.4 | 50.3 |
| OPT350M→FlanT5base | 1.1 | 122.1 | 34.0 | 49.9 |
| OPT2.7B→FlanT5xL | 106.5 | 133.2 | 51.3 | 53.5 |
| Transfer | COCO Caption | VQAv2 | ||
| linear | best | linear | best | |
| FlanT5base → OPT350M | 110.8 | 136.5 | 40.4 | 44.2 |
| FlanT5xL →OPT2.7B | 132.1 | 139.3 | 50.4 | 50.3 |
| OPT350M→FlanT5base | 1.1 | 122.1 | 34.0 | 49.9 |
| OPT2.7B→FlanT5xL | 106.5 | 133.2 | 51.3 | 53.5 |
| Graph G | RicG |
| C3 Cycles C4 Cn≥5 | 3 |
| 1 | |
| 0 | |
| n | |
| Complete Kn Grid Gn | n-1 0 |
| Tree Tr | 4 r+1-2 |
| H(G) | Cornell 0.11 | Texas 0.06 | Wisconsin 0.16 | Chameleon 0.25 | Squirrel 0.22 | Actor 0.24 | Cora 0.83 | Citeseer 0.71 | Pubmed 0.79 |
| None | 52.69 ± 0.21 | 61.19 ± 0.49 | 54.60±0.86 | 41.80 ± 0.41 | 39.83±0.14 | 28.70±0.09 | 81.89 ±0.79 | 72.31 ±0.17 | 78.16 ±0.23 |
| Undirected | 53.20 ±0.53 | 63.38 ±0.87 | 51.37 ± 1.15 | 42.63±0.30 | 40.77 ± 0.16 | 28.10 ±0.11 | = | ||
| +FA | 58.29±0.4964.82±0.29 | 55.48±0.62 | 42.33± 0.17 | 40.74 ± 0.13 | 28.68±0.16 | 81.65±0.18 | 70.47±0.18 79.48±0.12 | ||
| DIGL (PPR) | 58.26±0.50 | 62.03 ±0.43 | 49.53±0.27 | 42.02 ±0.13 | 34.38 ±0.11 | 30.79±0.10 83.21±0.27 73.29±0.17 78.84±0.08 | |||
| DIGL + Undirected 59.54 ±0.64 | 63.54±0.38 | 52.23±0.54 | 42.68± 0.12 | 33.36±0.21 | 29.71 ± 0.11 | = | |||
| SDRF | 54.60 ±0.39 | 64.46±0.38 | 55.51±0.27 43.75±0.31 40.97±0.14 29.70 ±0.13 82.76±0.23 72.58±0.20 79.10 ±0.11 | ||||||
| SDRF+ Undirected 57.54±0.34 70.35±0.60 61.55±0.86 44.46±0.17 41.47±0.21 29.85±0.07 |
| DIGL | SDRF | |
| Cornell | 351.1% /0.0% | 7.8% /33.3% |
| Texas | 483.3% /0.0% | 2.4% /10.4% |
| Wisconsin | 300.6%/0.0% | 1.4% / 7.5% |
| Chameleon | 336.1%/11.8% | 6.4% /6.4% |
| Squirrel | 228.8% /1.9% | 0.7%/0.7% |
| Actor | 2444.0% /2.3% | 5.4%/9.9% |
| Cora | 3038.0% /0.5% | 1.0% /1.0% |
| Citeseer | 2568.3%/ 0.0% | 1.1% / 1.1% |
| Pubmed | 2747.1% /0.1% | 0.2%/0.2% |
| Cornell | Texas | Wisconsin | Chameleon | Squirrel | Actor | Cora | Citeseer | Pubmed | |
| H(G) | 0.11 | 0.06 | 0.16 | 0.25 | 0.22 | 0.24 | 0.83 | 0.72 | 0.79 |
| Nodes | 140 | 135 | 184 | 832 | 2186 | 4388 | 2485 | 2120 | 19717 |
| Edges | 219 | 251 | 362 | 12355 | 65224 | 21907 | 5069 | 3679 | 44324 |
| Features | 1703 | 1703 | 1703 | 2323 | 2089 | 931 | 1433 | 3703 | 500 |
| Classes | 5 | 5 | 5 | 5 | 5 | 5 | 7 | 6 | 3 |
| Directed? | Yes | Yes | Yes | Yes | Yes | Yes | No | No | No |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay |
| Cornell | 0.3060 | 1 | 128 | 0.0082 | 0.1570 |
| Texas | 0.2346 | 1 | 128 | 0.0072 | 0.0037 |
| Wisconsin | 0.2869 | 1 | 64 | 0.0281 | 0.0113 |
| Chameleon | 0.7304 | 3 | 128 | 0.0248 | 0.0936 |
| Squirrel | 0.5974 | 2 | 64 | 0.0136 | 0.1346 |
| Actor | 0.7605 | 1 | 64 | 0.0290 | 0.0619 |
| Cora | 0.4144 | 1 | 64 | 0.0097 | 0.0639 |
| Citeseer | 0.7477 | 1 | 128 | 0.0251 | 0.4577 |
| Pubmed | 0.4013 | 1 | 64 | 0.0095 | 0.0448 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay |
| Cornell | 0.6910 | 1 | 64 | 0.0185 | 0.0285 |
| Texas | 0.2665 | 1 | 128 | 0.0069 | 0.0035 |
| Wisconsin | 0.2893 | 2 | 128 | 0.0142 | 0.0001 |
| Chameleon | 0.4657 | 3 | 64 | 0.0189 | 0.0423 |
| Squirrel | 0.5944 | 2 | 64 | 0.0081 | 0.0309 |
| Actor | 0.6626 | 2 | 64 | 0.0195 | 0.0219 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay |
| Cornell | 0.2643 | 1 | 128 | 0.0216 | 0.0760 |
| Texas | 0.2207 | 1 | 128 | 0.0102 | 0.4450 |
| Wisconsin | 0.2613 | 3 | 64 | 0.0057 | 0.0131 |
| Chameleon | 0.7783 | 3 | 64 | 0.0156 | 0.0108 |
| Squirrel | 0.3654 | 3 | 64 | 0.0077 | 0.1922 |
| Actor | 0.3824 | 1 | 64 | 0.0165 | 0.1168 |
| Cora | 0.7840 | 2 | 128 | 0.0149 | 0.1429 |
| Citeseer | 0.5460 | 2 | 64 | 0.0066 | 0.0758 |
| Pubmed | 0.3376 | 2 | 128 | 0.0204 | 0.0215 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | α | k | E |
| Cornell | 0.6294 | 1 | 64 | 0.0134 | 0.0258 | 0.1795 | 64 | - |
| Texas | 0.2382 | 2 | 128 | 0.0063 | 0.0153 | 0.0206 | 32 | - |
| Wisconsin | 0.2941 | 1 | 128 | 0.0083 | 0.0226 | 0.1246 | 1 | 0.0001 |
| Chameleon | 0.4191 | 1 | 128 | 0.0052 | 0.0001 | 0.0244 | 64 | 1 |
| Squirrel | 0.6844 | 1 | 128 | 0.0056 | 0.4537 | 0.0395 | 32 | |
| Actor | 0.7820 | 1 | 64 | 0.0170 | 0.0102 | 0.1584 | 128 | 、 |
| Cora | 0.3315 | 1 | 64 | 0.0284 | 0.0572 | 0.0773 | 128 | 1 |
| Citeseer | 0.5561 | 1 | 64 | 0.0094 | 0.5013 | 0.1076 | - | 0.0008 |
| Pubmed | 0.4915 | 2 | 128 | 0.0057 | 0.0597 | 0.1155 | 128 | - |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | α | k | E |
| Cornell | 0.6294 | 1 | 64 | 0.0134 | 0.0258 | 0.1795 | 64 | = |
| Texas | 0.2382 | 2 | 128 | 0.0063 | 0.0153 | 0.0206 | 32 | 1 |
| Wisconsin | 0.2941 | 1 | 128 | 0.0083 | 0.0226 | 0.1246 | 1 | 0.0001 |
| Chameleon | 0.4191 | 1 | 128 | 0.0052 | 0.0001 | 0.0244 | 64 | 1 |
| Squirrel | 0.7094 | 1 | 64 | 0.0172 | 0.0192 | 0.1610 | 64 | 1 |
| Actor | 0.4012 | 1 | 64 | 0.0161 | 0.0141 | 0.0706 | 1 | 0.0016 |
| Cora | 0.3315 | 1 | 64 | 0.0284 | 0.0572 | 0.0773 | 128 | - |
| Citeseer | 0.5561 | 1 | 64 | 0.0094 | 0.5013 | 0.1076 | - | 0.0008 |
| Pubmed | 0.4915 | 2 | 128 | 0.0057 | 0.0597 | 0.1155 | 128 | 1 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | Max iterations | T | C+ |
| Cornell | 0.2411 | 1 | 128 | 0.0172 | 0.0125 | 135 | 130 | 0.25 |
| Texas | 0.5954 | 1 | 128 | 0.0278 | 0.0623 | 47 | 172 | 2.25 |
| Wisconsin | 0.6033 | 1 | 128 | 0.0295 | 0.1920 | 27 | 32 | 0.5 |
| Chameleon | 0.7265 | 1 | 128 | 0.0180 | 0.2101 | 832 | 77 | 3.35 |
| Squirrel | 0.7401 | 2 | 16 | 0.0189 | 0.2255 | 6157 | 178 | 0.5 |
| Actor | 0.6886 | 1 | 128 | 0.0095 | 0.0727 | 1010 | 69 | 1.22 |
| Cora | 0.3396 | 1 | 128 | 0.0244 | 0.1076 | 100 | 163 | 0.95 |
| Citeseer | 0.4103 | 1 | 64 | 0.0199 | 0.4551 | 84 | 180 | 0.22 |
| Pubmed | 0.3749 | 3 | 128 | 0.0112 | 0.0138 | 166 | 115 | 14.43 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | Max iterations | T | C+ |
| Cornell | 0.2911 | 1 | 128 | 0.0056 | 0.0336 | 126 | 145 | 0.88 |
| Texas | 0.2160 | 1 | 64 | 0.0229 | 0.0137 | 89 | 22 | 1.64 |
| Wisconsin | 0.2452 | 1 | 64 | 0.0113 | 0.1559 | 136 | 12 | 7.95 |
| Chameleon | 0.4886 | 1 | 32 | 0.0268 | 0.4056 | 2441 | 252 | 2.84 |
| Squirrel | 0.4249 | 1 | 64 | 0.0295 | 0.1397 | 787 | 43 | 17.19 |
| Actor | 0.6705 | 1 | 128 | 0.0115 | 0.0447 | 1141 | 44 | 11.17 |
| Component | Specification |
| Architecture | x86_64 |
| CPU | 40x Intel(R) Xeon(R) Silver 4210R CPU @ 2.40GHz |
| GPU | 4x GeForce RTX 3090 (24268MiB/GPU) |
| RAM | 126GB |
| OS | Ubuntu 20.04.2LTS |
| Graph G | RicG |
| C3 Cycles C4 Cn≥5 | 3 |
| 1 | |
| 0 | |
| n | |
| Complete Kn Grid Gn | n-1 0 |
| Tree Tr | 4 r+1-2 |
| H(G) | Cornell 0.11 | Texas 0.06 | Wisconsin 0.16 | Chameleon 0.25 | Squirrel 0.22 | Actor 0.24 | Cora 0.83 | Citeseer 0.71 | Pubmed 0.79 |
| None | 52.69 ± 0.21 | 61.19 ± 0.49 | 54.60±0.86 | 41.80 ± 0.41 | 39.83±0.14 | 28.70±0.09 | 81.89 ±0.79 | 72.31 ±0.17 | 78.16 ±0.23 |
| Undirected | 53.20 ±0.53 | 63.38 ±0.87 | 51.37 ± 1.15 | 42.63±0.30 | 40.77 ± 0.16 | 28.10 ±0.11 | = | ||
| +FA | 58.29±0.4964.82±0.29 | 55.48±0.62 | 42.33± 0.17 | 40.74 ± 0.13 | 28.68±0.16 | 81.65±0.18 | 70.47±0.18 79.48±0.12 | ||
| DIGL (PPR) | 58.26±0.50 | 62.03 ±0.43 | 49.53±0.27 | 42.02 ±0.13 | 34.38 ±0.11 | 30.79±0.10 83.21±0.27 73.29±0.17 78.84±0.08 | |||
| DIGL + Undirected 59.54 ±0.64 | 63.54±0.38 | 52.23±0.54 | 42.68± 0.12 | 33.36±0.21 | 29.71 ± 0.11 | = | |||
| SDRF | 54.60 ±0.39 | 64.46±0.38 | 55.51±0.27 43.75±0.31 40.97±0.14 29.70 ±0.13 82.76±0.23 72.58±0.20 79.10 ±0.11 | ||||||
| SDRF+ Undirected 57.54±0.34 70.35±0.60 61.55±0.86 44.46±0.17 41.47±0.21 29.85±0.07 |
| DIGL | SDRF | |
| Cornell | 351.1% /0.0% | 7.8% /33.3% |
| Texas | 483.3% /0.0% | 2.4% /10.4% |
| Wisconsin | 300.6%/0.0% | 1.4% / 7.5% |
| Chameleon | 336.1%/11.8% | 6.4% /6.4% |
| Squirrel | 228.8% /1.9% | 0.7%/0.7% |
| Actor | 2444.0% /2.3% | 5.4%/9.9% |
| Cora | 3038.0% /0.5% | 1.0% /1.0% |
| Citeseer | 2568.3%/ 0.0% | 1.1% / 1.1% |
| Pubmed | 2747.1% /0.1% | 0.2%/0.2% |
| Cornell | Texas | Wisconsin | Chameleon | Squirrel | Actor | Cora | Citeseer | Pubmed | |
| H(G) | 0.11 | 0.06 | 0.16 | 0.25 | 0.22 | 0.24 | 0.83 | 0.72 | 0.79 |
| Nodes | 140 | 135 | 184 | 832 | 2186 | 4388 | 2485 | 2120 | 19717 |
| Edges | 219 | 251 | 362 | 12355 | 65224 | 21907 | 5069 | 3679 | 44324 |
| Features | 1703 | 1703 | 1703 | 2323 | 2089 | 931 | 1433 | 3703 | 500 |
| Classes | 5 | 5 | 5 | 5 | 5 | 5 | 7 | 6 | 3 |
| Directed? | Yes | Yes | Yes | Yes | Yes | Yes | No | No | No |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay |
| Cornell | 0.3060 | 1 | 128 | 0.0082 | 0.1570 |
| Texas | 0.2346 | 1 | 128 | 0.0072 | 0.0037 |
| Wisconsin | 0.2869 | 1 | 64 | 0.0281 | 0.0113 |
| Chameleon | 0.7304 | 3 | 128 | 0.0248 | 0.0936 |
| Squirrel | 0.5974 | 2 | 64 | 0.0136 | 0.1346 |
| Actor | 0.7605 | 1 | 64 | 0.0290 | 0.0619 |
| Cora | 0.4144 | 1 | 64 | 0.0097 | 0.0639 |
| Citeseer | 0.7477 | 1 | 128 | 0.0251 | 0.4577 |
| Pubmed | 0.4013 | 1 | 64 | 0.0095 | 0.0448 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay |
| Cornell | 0.6910 | 1 | 64 | 0.0185 | 0.0285 |
| Texas | 0.2665 | 1 | 128 | 0.0069 | 0.0035 |
| Wisconsin | 0.2893 | 2 | 128 | 0.0142 | 0.0001 |
| Chameleon | 0.4657 | 3 | 64 | 0.0189 | 0.0423 |
| Squirrel | 0.5944 | 2 | 64 | 0.0081 | 0.0309 |
| Actor | 0.6626 | 2 | 64 | 0.0195 | 0.0219 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay |
| Cornell | 0.2643 | 1 | 128 | 0.0216 | 0.0760 |
| Texas | 0.2207 | 1 | 128 | 0.0102 | 0.4450 |
| Wisconsin | 0.2613 | 3 | 64 | 0.0057 | 0.0131 |
| Chameleon | 0.7783 | 3 | 64 | 0.0156 | 0.0108 |
| Squirrel | 0.3654 | 3 | 64 | 0.0077 | 0.1922 |
| Actor | 0.3824 | 1 | 64 | 0.0165 | 0.1168 |
| Cora | 0.7840 | 2 | 128 | 0.0149 | 0.1429 |
| Citeseer | 0.5460 | 2 | 64 | 0.0066 | 0.0758 |
| Pubmed | 0.3376 | 2 | 128 | 0.0204 | 0.0215 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | α | k | E |
| Cornell | 0.6294 | 1 | 64 | 0.0134 | 0.0258 | 0.1795 | 64 | - |
| Texas | 0.2382 | 2 | 128 | 0.0063 | 0.0153 | 0.0206 | 32 | - |
| Wisconsin | 0.2941 | 1 | 128 | 0.0083 | 0.0226 | 0.1246 | 1 | 0.0001 |
| Chameleon | 0.4191 | 1 | 128 | 0.0052 | 0.0001 | 0.0244 | 64 | 1 |
| Squirrel | 0.6844 | 1 | 128 | 0.0056 | 0.4537 | 0.0395 | 32 | |
| Actor | 0.7820 | 1 | 64 | 0.0170 | 0.0102 | 0.1584 | 128 | 、 |
| Cora | 0.3315 | 1 | 64 | 0.0284 | 0.0572 | 0.0773 | 128 | 1 |
| Citeseer | 0.5561 | 1 | 64 | 0.0094 | 0.5013 | 0.1076 | - | 0.0008 |
| Pubmed | 0.4915 | 2 | 128 | 0.0057 | 0.0597 | 0.1155 | 128 | - |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | α | k | E |
| Cornell | 0.6294 | 1 | 64 | 0.0134 | 0.0258 | 0.1795 | 64 | = |
| Texas | 0.2382 | 2 | 128 | 0.0063 | 0.0153 | 0.0206 | 32 | 1 |
| Wisconsin | 0.2941 | 1 | 128 | 0.0083 | 0.0226 | 0.1246 | 1 | 0.0001 |
| Chameleon | 0.4191 | 1 | 128 | 0.0052 | 0.0001 | 0.0244 | 64 | 1 |
| Squirrel | 0.7094 | 1 | 64 | 0.0172 | 0.0192 | 0.1610 | 64 | 1 |
| Actor | 0.4012 | 1 | 64 | 0.0161 | 0.0141 | 0.0706 | 1 | 0.0016 |
| Cora | 0.3315 | 1 | 64 | 0.0284 | 0.0572 | 0.0773 | 128 | - |
| Citeseer | 0.5561 | 1 | 64 | 0.0094 | 0.5013 | 0.1076 | - | 0.0008 |
| Pubmed | 0.4915 | 2 | 128 | 0.0057 | 0.0597 | 0.1155 | 128 | 1 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | Max iterations | T | C+ |
| Cornell | 0.2411 | 1 | 128 | 0.0172 | 0.0125 | 135 | 130 | 0.25 |
| Texas | 0.5954 | 1 | 128 | 0.0278 | 0.0623 | 47 | 172 | 2.25 |
| Wisconsin | 0.6033 | 1 | 128 | 0.0295 | 0.1920 | 27 | 32 | 0.5 |
| Chameleon | 0.7265 | 1 | 128 | 0.0180 | 0.2101 | 832 | 77 | 3.35 |
| Squirrel | 0.7401 | 2 | 16 | 0.0189 | 0.2255 | 6157 | 178 | 0.5 |
| Actor | 0.6886 | 1 | 128 | 0.0095 | 0.0727 | 1010 | 69 | 1.22 |
| Cora | 0.3396 | 1 | 128 | 0.0244 | 0.1076 | 100 | 163 | 0.95 |
| Citeseer | 0.4103 | 1 | 64 | 0.0199 | 0.4551 | 84 | 180 | 0.22 |
| Pubmed | 0.3749 | 3 | 128 | 0.0112 | 0.0138 | 166 | 115 | 14.43 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | Max iterations | T | C+ |
| Cornell | 0.2911 | 1 | 128 | 0.0056 | 0.0336 | 126 | 145 | 0.88 |
| Texas | 0.2160 | 1 | 64 | 0.0229 | 0.0137 | 89 | 22 | 1.64 |
| Wisconsin | 0.2452 | 1 | 64 | 0.0113 | 0.1559 | 136 | 12 | 7.95 |
| Chameleon | 0.4886 | 1 | 32 | 0.0268 | 0.4056 | 2441 | 252 | 2.84 |
| Squirrel | 0.4249 | 1 | 64 | 0.0295 | 0.1397 | 787 | 43 | 17.19 |
| Actor | 0.6705 | 1 | 128 | 0.0115 | 0.0447 | 1141 | 44 | 11.17 |
| Component | Specification |
| Architecture | x86_64 |
| CPU | 40x Intel(R) Xeon(R) Silver 4210R CPU @ 2.40GHz |
| GPU | 4x GeForce RTX 3090 (24268MiB/GPU) |
| RAM | 126GB |
| OS | Ubuntu 20.04.2LTS |
| Graph G | RicG |
| C3 Cycles C4 Cn≥5 | 3 |
| 1 | |
| 0 | |
| n | |
| Complete Kn Grid Gn | n-1 0 |
| Tree Tr | 4 r+1-2 |
| Graph G | RicG |
| C3 Cycles C4 Cn≥5 | 3 |
| 1 | |
| 0 | |
| n | |
| Complete Kn Grid Gn | n-1 0 |
| Tree Tr | 4 r+1-2 |
| H(G) | Cornell 0.11 | Texas 0.06 | Wisconsin 0.16 | Chameleon 0.25 | Squirrel 0.22 | Actor 0.24 | Cora 0.83 | Citeseer 0.71 | Pubmed 0.79 |
| None | 52.69 ± 0.21 | 61.19 ± 0.49 | 54.60±0.86 | 41.80 ± 0.41 | 39.83±0.14 | 28.70±0.09 | 81.89 ±0.79 | 72.31 ±0.17 | 78.16 ±0.23 |
| Undirected | 53.20 ±0.53 | 63.38 ±0.87 | 51.37 ± 1.15 | 42.63±0.30 | 40.77 ± 0.16 | 28.10 ±0.11 | = | ||
| +FA | 58.29±0.4964.82±0.29 | 55.48±0.62 | 42.33± 0.17 | 40.74 ± 0.13 | 28.68±0.16 | 81.65±0.18 | 70.47±0.18 79.48±0.12 | ||
| DIGL (PPR) | 58.26±0.50 | 62.03 ±0.43 | 49.53±0.27 | 42.02 ±0.13 | 34.38 ±0.11 | 30.79±0.10 83.21±0.27 73.29±0.17 78.84±0.08 | |||
| DIGL + Undirected 59.54 ±0.64 | 63.54±0.38 | 52.23±0.54 | 42.68± 0.12 | 33.36±0.21 | 29.71 ± 0.11 | = | |||
| SDRF | 54.60 ±0.39 | 64.46±0.38 | 55.51±0.27 43.75±0.31 40.97±0.14 29.70 ±0.13 82.76±0.23 72.58±0.20 79.10 ±0.11 | ||||||
| SDRF+ Undirected 57.54±0.34 70.35±0.60 61.55±0.86 44.46±0.17 41.47±0.21 29.85±0.07 |
| H(G) | Cornell 0.11 | Texas 0.06 | Wisconsin 0.16 | Chameleon 0.25 | Squirrel 0.22 | Actor 0.24 | Cora 0.83 | Citeseer 0.71 | Pubmed 0.79 |
| None | 52.69 ± 0.21 | 61.19 ± 0.49 | 54.60±0.86 | 41.80 ± 0.41 | 39.83±0.14 | 28.70±0.09 | 81.89 ±0.79 | 72.31 ±0.17 | 78.16 ±0.23 |
| Undirected | 53.20 ±0.53 | 63.38 ±0.87 | 51.37 ± 1.15 | 42.63±0.30 | 40.77 ± 0.16 | 28.10 ±0.11 | = | ||
| +FA | 58.29±0.4964.82±0.29 | 55.48±0.62 | 42.33± 0.17 | 40.74 ± 0.13 | 28.68±0.16 | 81.65±0.18 | 70.47±0.18 79.48±0.12 | ||
| DIGL (PPR) | 58.26±0.50 | 62.03 ±0.43 | 49.53±0.27 | 42.02 ±0.13 | 34.38 ±0.11 | 30.79±0.10 83.21±0.27 73.29±0.17 78.84±0.08 | |||
| DIGL + Undirected 59.54 ±0.64 | 63.54±0.38 | 52.23±0.54 | 42.68± 0.12 | 33.36±0.21 | 29.71 ± 0.11 | = | |||
| SDRF | 54.60 ±0.39 | 64.46±0.38 | 55.51±0.27 43.75±0.31 40.97±0.14 29.70 ±0.13 82.76±0.23 72.58±0.20 79.10 ±0.11 | ||||||
| SDRF+ Undirected 57.54±0.34 70.35±0.60 61.55±0.86 44.46±0.17 41.47±0.21 29.85±0.07 |
| DIGL | SDRF | |
| Cornell | 351.1% /0.0% | 7.8% /33.3% |
| Texas | 483.3% /0.0% | 2.4% /10.4% |
| Wisconsin | 300.6%/0.0% | 1.4% / 7.5% |
| Chameleon | 336.1%/11.8% | 6.4% /6.4% |
| Squirrel | 228.8% /1.9% | 0.7%/0.7% |
| Actor | 2444.0% /2.3% | 5.4%/9.9% |
| Cora | 3038.0% /0.5% | 1.0% /1.0% |
| Citeseer | 2568.3%/ 0.0% | 1.1% / 1.1% |
| Pubmed | 2747.1% /0.1% | 0.2%/0.2% |
| DIGL | SDRF | |
| Cornell | 351.1% /0.0% | 7.8% /33.3% |
| Texas | 483.3% /0.0% | 2.4% /10.4% |
| Wisconsin | 300.6%/0.0% | 1.4% / 7.5% |
| Chameleon | 336.1%/11.8% | 6.4% /6.4% |
| Squirrel | 228.8% /1.9% | 0.7%/0.7% |
| Actor | 2444.0% /2.3% | 5.4%/9.9% |
| Cora | 3038.0% /0.5% | 1.0% /1.0% |
| Citeseer | 2568.3%/ 0.0% | 1.1% / 1.1% |
| Pubmed | 2747.1% /0.1% | 0.2%/0.2% |
| Cornell | Texas | Wisconsin | Chameleon | Squirrel | Actor | Cora | Citeseer | Pubmed | |
| H(G) | 0.11 | 0.06 | 0.16 | 0.25 | 0.22 | 0.24 | 0.83 | 0.72 | 0.79 |
| Nodes | 140 | 135 | 184 | 832 | 2186 | 4388 | 2485 | 2120 | 19717 |
| Edges | 219 | 251 | 362 | 12355 | 65224 | 21907 | 5069 | 3679 | 44324 |
| Features | 1703 | 1703 | 1703 | 2323 | 2089 | 931 | 1433 | 3703 | 500 |
| Classes | 5 | 5 | 5 | 5 | 5 | 5 | 7 | 6 | 3 |
| Directed? | Yes | Yes | Yes | Yes | Yes | Yes | No | No | No |
| Cornell | Texas | Wisconsin | Chameleon | Squirrel | Actor | Cora | Citeseer | Pubmed | |
| H(G) | 0.11 | 0.06 | 0.16 | 0.25 | 0.22 | 0.24 | 0.83 | 0.72 | 0.79 |
| Nodes | 140 | 135 | 184 | 832 | 2186 | 4388 | 2485 | 2120 | 19717 |
| Edges | 219 | 251 | 362 | 12355 | 65224 | 21907 | 5069 | 3679 | 44324 |
| Features | 1703 | 1703 | 1703 | 2323 | 2089 | 931 | 1433 | 3703 | 500 |
| Classes | 5 | 5 | 5 | 5 | 5 | 5 | 7 | 6 | 3 |
| Directed? | Yes | Yes | Yes | Yes | Yes | Yes | No | No | No |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay |
| Cornell | 0.3060 | 1 | 128 | 0.0082 | 0.1570 |
| Texas | 0.2346 | 1 | 128 | 0.0072 | 0.0037 |
| Wisconsin | 0.2869 | 1 | 64 | 0.0281 | 0.0113 |
| Chameleon | 0.7304 | 3 | 128 | 0.0248 | 0.0936 |
| Squirrel | 0.5974 | 2 | 64 | 0.0136 | 0.1346 |
| Actor | 0.7605 | 1 | 64 | 0.0290 | 0.0619 |
| Cora | 0.4144 | 1 | 64 | 0.0097 | 0.0639 |
| Citeseer | 0.7477 | 1 | 128 | 0.0251 | 0.4577 |
| Pubmed | 0.4013 | 1 | 64 | 0.0095 | 0.0448 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay |
| Cornell | 0.6910 | 1 | 64 | 0.0185 | 0.0285 |
| Texas | 0.2665 | 1 | 128 | 0.0069 | 0.0035 |
| Wisconsin | 0.2893 | 2 | 128 | 0.0142 | 0.0001 |
| Chameleon | 0.4657 | 3 | 64 | 0.0189 | 0.0423 |
| Squirrel | 0.5944 | 2 | 64 | 0.0081 | 0.0309 |
| Actor | 0.6626 | 2 | 64 | 0.0195 | 0.0219 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay |
| Cornell | 0.3060 | 1 | 128 | 0.0082 | 0.1570 |
| Texas | 0.2346 | 1 | 128 | 0.0072 | 0.0037 |
| Wisconsin | 0.2869 | 1 | 64 | 0.0281 | 0.0113 |
| Chameleon | 0.7304 | 3 | 128 | 0.0248 | 0.0936 |
| Squirrel | 0.5974 | 2 | 64 | 0.0136 | 0.1346 |
| Actor | 0.7605 | 1 | 64 | 0.0290 | 0.0619 |
| Cora | 0.4144 | 1 | 64 | 0.0097 | 0.0639 |
| Citeseer | 0.7477 | 1 | 128 | 0.0251 | 0.4577 |
| Pubmed | 0.4013 | 1 | 64 | 0.0095 | 0.0448 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay |
| Cornell | 0.6910 | 1 | 64 | 0.0185 | 0.0285 |
| Texas | 0.2665 | 1 | 128 | 0.0069 | 0.0035 |
| Wisconsin | 0.2893 | 2 | 128 | 0.0142 | 0.0001 |
| Chameleon | 0.4657 | 3 | 64 | 0.0189 | 0.0423 |
| Squirrel | 0.5944 | 2 | 64 | 0.0081 | 0.0309 |
| Actor | 0.6626 | 2 | 64 | 0.0195 | 0.0219 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay |
| Cornell | 0.2643 | 1 | 128 | 0.0216 | 0.0760 |
| Texas | 0.2207 | 1 | 128 | 0.0102 | 0.4450 |
| Wisconsin | 0.2613 | 3 | 64 | 0.0057 | 0.0131 |
| Chameleon | 0.7783 | 3 | 64 | 0.0156 | 0.0108 |
| Squirrel | 0.3654 | 3 | 64 | 0.0077 | 0.1922 |
| Actor | 0.3824 | 1 | 64 | 0.0165 | 0.1168 |
| Cora | 0.7840 | 2 | 128 | 0.0149 | 0.1429 |
| Citeseer | 0.5460 | 2 | 64 | 0.0066 | 0.0758 |
| Pubmed | 0.3376 | 2 | 128 | 0.0204 | 0.0215 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | α | k | E |
| Cornell | 0.6294 | 1 | 64 | 0.0134 | 0.0258 | 0.1795 | 64 | - |
| Texas | 0.2382 | 2 | 128 | 0.0063 | 0.0153 | 0.0206 | 32 | - |
| Wisconsin | 0.2941 | 1 | 128 | 0.0083 | 0.0226 | 0.1246 | 1 | 0.0001 |
| Chameleon | 0.4191 | 1 | 128 | 0.0052 | 0.0001 | 0.0244 | 64 | 1 |
| Squirrel | 0.6844 | 1 | 128 | 0.0056 | 0.4537 | 0.0395 | 32 | |
| Actor | 0.7820 | 1 | 64 | 0.0170 | 0.0102 | 0.1584 | 128 | 、 |
| Cora | 0.3315 | 1 | 64 | 0.0284 | 0.0572 | 0.0773 | 128 | 1 |
| Citeseer | 0.5561 | 1 | 64 | 0.0094 | 0.5013 | 0.1076 | - | 0.0008 |
| Pubmed | 0.4915 | 2 | 128 | 0.0057 | 0.0597 | 0.1155 | 128 | - |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | α | k | E |
| Cornell | 0.6294 | 1 | 64 | 0.0134 | 0.0258 | 0.1795 | 64 | = |
| Texas | 0.2382 | 2 | 128 | 0.0063 | 0.0153 | 0.0206 | 32 | 1 |
| Wisconsin | 0.2941 | 1 | 128 | 0.0083 | 0.0226 | 0.1246 | 1 | 0.0001 |
| Chameleon | 0.4191 | 1 | 128 | 0.0052 | 0.0001 | 0.0244 | 64 | 1 |
| Squirrel | 0.7094 | 1 | 64 | 0.0172 | 0.0192 | 0.1610 | 64 | 1 |
| Actor | 0.4012 | 1 | 64 | 0.0161 | 0.0141 | 0.0706 | 1 | 0.0016 |
| Cora | 0.3315 | 1 | 64 | 0.0284 | 0.0572 | 0.0773 | 128 | - |
| Citeseer | 0.5561 | 1 | 64 | 0.0094 | 0.5013 | 0.1076 | - | 0.0008 |
| Pubmed | 0.4915 | 2 | 128 | 0.0057 | 0.0597 | 0.1155 | 128 | 1 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay |
| Cornell | 0.2643 | 1 | 128 | 0.0216 | 0.0760 |
| Texas | 0.2207 | 1 | 128 | 0.0102 | 0.4450 |
| Wisconsin | 0.2613 | 3 | 64 | 0.0057 | 0.0131 |
| Chameleon | 0.7783 | 3 | 64 | 0.0156 | 0.0108 |
| Squirrel | 0.3654 | 3 | 64 | 0.0077 | 0.1922 |
| Actor | 0.3824 | 1 | 64 | 0.0165 | 0.1168 |
| Cora | 0.7840 | 2 | 128 | 0.0149 | 0.1429 |
| Citeseer | 0.5460 | 2 | 64 | 0.0066 | 0.0758 |
| Pubmed | 0.3376 | 2 | 128 | 0.0204 | 0.0215 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | α | k | E |
| Cornell | 0.6294 | 1 | 64 | 0.0134 | 0.0258 | 0.1795 | 64 | - |
| Texas | 0.2382 | 2 | 128 | 0.0063 | 0.0153 | 0.0206 | 32 | - |
| Wisconsin | 0.2941 | 1 | 128 | 0.0083 | 0.0226 | 0.1246 | 1 | 0.0001 |
| Chameleon | 0.4191 | 1 | 128 | 0.0052 | 0.0001 | 0.0244 | 64 | 1 |
| Squirrel | 0.6844 | 1 | 128 | 0.0056 | 0.4537 | 0.0395 | 32 | |
| Actor | 0.7820 | 1 | 64 | 0.0170 | 0.0102 | 0.1584 | 128 | 、 |
| Cora | 0.3315 | 1 | 64 | 0.0284 | 0.0572 | 0.0773 | 128 | 1 |
| Citeseer | 0.5561 | 1 | 64 | 0.0094 | 0.5013 | 0.1076 | - | 0.0008 |
| Pubmed | 0.4915 | 2 | 128 | 0.0057 | 0.0597 | 0.1155 | 128 | - |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | α | k | E |
| Cornell | 0.6294 | 1 | 64 | 0.0134 | 0.0258 | 0.1795 | 64 | = |
| Texas | 0.2382 | 2 | 128 | 0.0063 | 0.0153 | 0.0206 | 32 | 1 |
| Wisconsin | 0.2941 | 1 | 128 | 0.0083 | 0.0226 | 0.1246 | 1 | 0.0001 |
| Chameleon | 0.4191 | 1 | 128 | 0.0052 | 0.0001 | 0.0244 | 64 | 1 |
| Squirrel | 0.7094 | 1 | 64 | 0.0172 | 0.0192 | 0.1610 | 64 | 1 |
| Actor | 0.4012 | 1 | 64 | 0.0161 | 0.0141 | 0.0706 | 1 | 0.0016 |
| Cora | 0.3315 | 1 | 64 | 0.0284 | 0.0572 | 0.0773 | 128 | - |
| Citeseer | 0.5561 | 1 | 64 | 0.0094 | 0.5013 | 0.1076 | - | 0.0008 |
| Pubmed | 0.4915 | 2 | 128 | 0.0057 | 0.0597 | 0.1155 | 128 | 1 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | Max iterations | T | C+ |
| Cornell | 0.2411 | 1 | 128 | 0.0172 | 0.0125 | 135 | 130 | 0.25 |
| Texas | 0.5954 | 1 | 128 | 0.0278 | 0.0623 | 47 | 172 | 2.25 |
| Wisconsin | 0.6033 | 1 | 128 | 0.0295 | 0.1920 | 27 | 32 | 0.5 |
| Chameleon | 0.7265 | 1 | 128 | 0.0180 | 0.2101 | 832 | 77 | 3.35 |
| Squirrel | 0.7401 | 2 | 16 | 0.0189 | 0.2255 | 6157 | 178 | 0.5 |
| Actor | 0.6886 | 1 | 128 | 0.0095 | 0.0727 | 1010 | 69 | 1.22 |
| Cora | 0.3396 | 1 | 128 | 0.0244 | 0.1076 | 100 | 163 | 0.95 |
| Citeseer | 0.4103 | 1 | 64 | 0.0199 | 0.4551 | 84 | 180 | 0.22 |
| Pubmed | 0.3749 | 3 | 128 | 0.0112 | 0.0138 | 166 | 115 | 14.43 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | Max iterations | T | C+ |
| Cornell | 0.2911 | 1 | 128 | 0.0056 | 0.0336 | 126 | 145 | 0.88 |
| Texas | 0.2160 | 1 | 64 | 0.0229 | 0.0137 | 89 | 22 | 1.64 |
| Wisconsin | 0.2452 | 1 | 64 | 0.0113 | 0.1559 | 136 | 12 | 7.95 |
| Chameleon | 0.4886 | 1 | 32 | 0.0268 | 0.4056 | 2441 | 252 | 2.84 |
| Squirrel | 0.4249 | 1 | 64 | 0.0295 | 0.1397 | 787 | 43 | 17.19 |
| Actor | 0.6705 | 1 | 128 | 0.0115 | 0.0447 | 1141 | 44 | 11.17 |
| Component | Specification |
| Architecture | x86_64 |
| CPU | 40x Intel(R) Xeon(R) Silver 4210R CPU @ 2.40GHz |
| GPU | 4x GeForce RTX 3090 (24268MiB/GPU) |
| RAM | 126GB |
| OS | Ubuntu 20.04.2LTS |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | Max iterations | T | C+ |
| Cornell | 0.2411 | 1 | 128 | 0.0172 | 0.0125 | 135 | 130 | 0.25 |
| Texas | 0.5954 | 1 | 128 | 0.0278 | 0.0623 | 47 | 172 | 2.25 |
| Wisconsin | 0.6033 | 1 | 128 | 0.0295 | 0.1920 | 27 | 32 | 0.5 |
| Chameleon | 0.7265 | 1 | 128 | 0.0180 | 0.2101 | 832 | 77 | 3.35 |
| Squirrel | 0.7401 | 2 | 16 | 0.0189 | 0.2255 | 6157 | 178 | 0.5 |
| Actor | 0.6886 | 1 | 128 | 0.0095 | 0.0727 | 1010 | 69 | 1.22 |
| Cora | 0.3396 | 1 | 128 | 0.0244 | 0.1076 | 100 | 163 | 0.95 |
| Citeseer | 0.4103 | 1 | 64 | 0.0199 | 0.4551 | 84 | 180 | 0.22 |
| Pubmed | 0.3749 | 3 | 128 | 0.0112 | 0.0138 | 166 | 115 | 14.43 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | Max iterations | T | C+ |
| Cornell | 0.2911 | 1 | 128 | 0.0056 | 0.0336 | 126 | 145 | 0.88 |
| Texas | 0.2160 | 1 | 64 | 0.0229 | 0.0137 | 89 | 22 | 1.64 |
| Wisconsin | 0.2452 | 1 | 64 | 0.0113 | 0.1559 | 136 | 12 | 7.95 |
| Chameleon | 0.4886 | 1 | 32 | 0.0268 | 0.4056 | 2441 | 252 | 2.84 |
| Squirrel | 0.4249 | 1 | 64 | 0.0295 | 0.1397 | 787 | 43 | 17.19 |
| Actor | 0.6705 | 1 | 128 | 0.0115 | 0.0447 | 1141 | 44 | 11.17 |
| Component | Specification |
| Architecture | x86_64 |
| CPU | 40x Intel(R) Xeon(R) Silver 4210R CPU @ 2.40GHz |
| GPU | 4x GeForce RTX 3090 (24268MiB/GPU) |
| RAM | 126GB |
| OS | Ubuntu 20.04.2LTS |
| Graph G | RicG |
| C3 Cycles C4 Cn≥5 | 3 |
| 1 | |
| 0 | |
| n | |
| Complete Kn Grid Gn | n-1 0 |
| Tree Tr | 4 r+1-2 |
| H(G) | Cornell 0.11 | Texas 0.06 | Wisconsin 0.16 | Chameleon 0.25 | Squirrel 0.22 | Actor 0.24 | Cora 0.83 | Citeseer 0.71 | Pubmed 0.79 |
| None | 52.69 ± 0.21 | 61.19 ± 0.49 | 54.60±0.86 | 41.80 ± 0.41 | 39.83±0.14 | 28.70±0.09 | 81.89 ±0.79 | 72.31 ±0.17 | 78.16 ±0.23 |
| Undirected | 53.20 ±0.53 | 63.38 ±0.87 | 51.37 ± 1.15 | 42.63±0.30 | 40.77 ± 0.16 | 28.10 ±0.11 | = | ||
| +FA | 58.29±0.4964.82±0.29 | 55.48±0.62 | 42.33± 0.17 | 40.74 ± 0.13 | 28.68±0.16 | 81.65±0.18 | 70.47±0.18 79.48±0.12 | ||
| DIGL (PPR) | 58.26±0.50 | 62.03 ±0.43 | 49.53±0.27 | 42.02 ±0.13 | 34.38 ±0.11 | 30.79±0.10 83.21±0.27 73.29±0.17 78.84±0.08 | |||
| DIGL + Undirected 59.54 ±0.64 | 63.54±0.38 | 52.23±0.54 | 42.68± 0.12 | 33.36±0.21 | 29.71 ± 0.11 | = | |||
| SDRF | 54.60 ±0.39 | 64.46±0.38 | 55.51±0.27 43.75±0.31 40.97±0.14 29.70 ±0.13 82.76±0.23 72.58±0.20 79.10 ±0.11 | ||||||
| SDRF+ Undirected 57.54±0.34 70.35±0.60 61.55±0.86 44.46±0.17 41.47±0.21 29.85±0.07 |
| DIGL | SDRF | |
| Cornell | 351.1% /0.0% | 7.8% /33.3% |
| Texas | 483.3% /0.0% | 2.4% /10.4% |
| Wisconsin | 300.6%/0.0% | 1.4% / 7.5% |
| Chameleon | 336.1%/11.8% | 6.4% /6.4% |
| Squirrel | 228.8% /1.9% | 0.7%/0.7% |
| Actor | 2444.0% /2.3% | 5.4%/9.9% |
| Cora | 3038.0% /0.5% | 1.0% /1.0% |
| Citeseer | 2568.3%/ 0.0% | 1.1% / 1.1% |
| Pubmed | 2747.1% /0.1% | 0.2%/0.2% |
| Cornell | Texas | Wisconsin | Chameleon | Squirrel | Actor | Cora | Citeseer | Pubmed | |
| H(G) | 0.11 | 0.06 | 0.16 | 0.25 | 0.22 | 0.24 | 0.83 | 0.72 | 0.79 |
| Nodes | 140 | 135 | 184 | 832 | 2186 | 4388 | 2485 | 2120 | 19717 |
| Edges | 219 | 251 | 362 | 12355 | 65224 | 21907 | 5069 | 3679 | 44324 |
| Features | 1703 | 1703 | 1703 | 2323 | 2089 | 931 | 1433 | 3703 | 500 |
| Classes | 5 | 5 | 5 | 5 | 5 | 5 | 7 | 6 | 3 |
| Directed? | Yes | Yes | Yes | Yes | Yes | Yes | No | No | No |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay |
| Cornell | 0.3060 | 1 | 128 | 0.0082 | 0.1570 |
| Texas | 0.2346 | 1 | 128 | 0.0072 | 0.0037 |
| Wisconsin | 0.2869 | 1 | 64 | 0.0281 | 0.0113 |
| Chameleon | 0.7304 | 3 | 128 | 0.0248 | 0.0936 |
| Squirrel | 0.5974 | 2 | 64 | 0.0136 | 0.1346 |
| Actor | 0.7605 | 1 | 64 | 0.0290 | 0.0619 |
| Cora | 0.4144 | 1 | 64 | 0.0097 | 0.0639 |
| Citeseer | 0.7477 | 1 | 128 | 0.0251 | 0.4577 |
| Pubmed | 0.4013 | 1 | 64 | 0.0095 | 0.0448 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay |
| Cornell | 0.6910 | 1 | 64 | 0.0185 | 0.0285 |
| Texas | 0.2665 | 1 | 128 | 0.0069 | 0.0035 |
| Wisconsin | 0.2893 | 2 | 128 | 0.0142 | 0.0001 |
| Chameleon | 0.4657 | 3 | 64 | 0.0189 | 0.0423 |
| Squirrel | 0.5944 | 2 | 64 | 0.0081 | 0.0309 |
| Actor | 0.6626 | 2 | 64 | 0.0195 | 0.0219 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | α | k | E |
| Cornell | 0.6294 | 1 | 64 | 0.0134 | 0.0258 | 0.1795 | 64 | - |
| Texas | 0.2382 | 2 | 128 | 0.0063 | 0.0153 | 0.0206 | 32 | - |
| Wisconsin | 0.2941 | 1 | 128 | 0.0083 | 0.0226 | 0.1246 | 1 | 0.0001 |
| Chameleon | 0.4191 | 1 | 128 | 0.0052 | 0.0001 | 0.0244 | 64 | 1 |
| Squirrel | 0.6844 | 1 | 128 | 0.0056 | 0.4537 | 0.0395 | 32 | |
| Actor | 0.7820 | 1 | 64 | 0.0170 | 0.0102 | 0.1584 | 128 | 、 |
| Cora | 0.3315 | 1 | 64 | 0.0284 | 0.0572 | 0.0773 | 128 | 1 |
| Citeseer | 0.5561 | 1 | 64 | 0.0094 | 0.5013 | 0.1076 | - | 0.0008 |
| Pubmed | 0.4915 | 2 | 128 | 0.0057 | 0.0597 | 0.1155 | 128 | - |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | α | k | E |
| Cornell | 0.6294 | 1 | 64 | 0.0134 | 0.0258 | 0.1795 | 64 | = |
| Texas | 0.2382 | 2 | 128 | 0.0063 | 0.0153 | 0.0206 | 32 | 1 |
| Wisconsin | 0.2941 | 1 | 128 | 0.0083 | 0.0226 | 0.1246 | 1 | 0.0001 |
| Chameleon | 0.4191 | 1 | 128 | 0.0052 | 0.0001 | 0.0244 | 64 | 1 |
| Squirrel | 0.7094 | 1 | 64 | 0.0172 | 0.0192 | 0.1610 | 64 | 1 |
| Actor | 0.4012 | 1 | 64 | 0.0161 | 0.0141 | 0.0706 | 1 | 0.0016 |
| Cora | 0.3315 | 1 | 64 | 0.0284 | 0.0572 | 0.0773 | 128 | - |
| Citeseer | 0.5561 | 1 | 64 | 0.0094 | 0.5013 | 0.1076 | - | 0.0008 |
| Pubmed | 0.4915 | 2 | 128 | 0.0057 | 0.0597 | 0.1155 | 128 | 1 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay |
| Cornell | 0.2643 | 1 | 128 | 0.0216 | 0.0760 |
| Texas | 0.2207 | 1 | 128 | 0.0102 | 0.4450 |
| Wisconsin | 0.2613 | 3 | 64 | 0.0057 | 0.0131 |
| Chameleon | 0.7783 | 3 | 64 | 0.0156 | 0.0108 |
| Squirrel | 0.3654 | 3 | 64 | 0.0077 | 0.1922 |
| Actor | 0.3824 | 1 | 64 | 0.0165 | 0.1168 |
| Cora | 0.7840 | 2 | 128 | 0.0149 | 0.1429 |
| Citeseer | 0.5460 | 2 | 64 | 0.0066 | 0.0758 |
| Pubmed | 0.3376 | 2 | 128 | 0.0204 | 0.0215 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | Max iterations | T | C+ |
| Cornell | 0.2411 | 1 | 128 | 0.0172 | 0.0125 | 135 | 130 | 0.25 |
| Texas | 0.5954 | 1 | 128 | 0.0278 | 0.0623 | 47 | 172 | 2.25 |
| Wisconsin | 0.6033 | 1 | 128 | 0.0295 | 0.1920 | 27 | 32 | 0.5 |
| Chameleon | 0.7265 | 1 | 128 | 0.0180 | 0.2101 | 832 | 77 | 3.35 |
| Squirrel | 0.7401 | 2 | 16 | 0.0189 | 0.2255 | 6157 | 178 | 0.5 |
| Actor | 0.6886 | 1 | 128 | 0.0095 | 0.0727 | 1010 | 69 | 1.22 |
| Cora | 0.3396 | 1 | 128 | 0.0244 | 0.1076 | 100 | 163 | 0.95 |
| Citeseer | 0.4103 | 1 | 64 | 0.0199 | 0.4551 | 84 | 180 | 0.22 |
| Pubmed | 0.3749 | 3 | 128 | 0.0112 | 0.0138 | 166 | 115 | 14.43 |
| Dataset | Dropout | Hidden depth | Hidden dimension | Learning rate | Weight decay | Max iterations | T | C+ |
| Cornell | 0.2911 | 1 | 128 | 0.0056 | 0.0336 | 126 | 145 | 0.88 |
| Texas | 0.2160 | 1 | 64 | 0.0229 | 0.0137 | 89 | 22 | 1.64 |
| Wisconsin | 0.2452 | 1 | 64 | 0.0113 | 0.1559 | 136 | 12 | 7.95 |
| Chameleon | 0.4886 | 1 | 32 | 0.0268 | 0.4056 | 2441 | 252 | 2.84 |
| Squirrel | 0.4249 | 1 | 64 | 0.0295 | 0.1397 | 787 | 43 | 17.19 |
| Actor | 0.6705 | 1 | 128 | 0.0115 | 0.0447 | 1141 | 44 | 11.17 |
| Component | Specification |
| Architecture | x86_64 |
| CPU | 40x Intel(R) Xeon(R) Silver 4210R CPU @ 2.40GHz |
| GPU | 4x GeForce RTX 3090 (24268MiB/GPU) |
| RAM | 126GB |
| OS | Ubuntu 20.04.2LTS |
| HINXS | Context: Question: Answer: | Christopher agrees with Kevin.Tifany agrees with Mathew.Mary hangs out with Danielle.James hangs out with Thomas.Kevin is a student.Matthew is a plumber.Danielle is a student. Thomas is a plumber. Who hangs out with a student? Explanation:Danielle is a student and Mary hangs out with Danielle. |
| Mary Premise: | A toddler in a green jersey is being folowed bya wheelchair bound woman in ared sweater past awooden bench. | |
| IINS-3 | Hypothesis: Label: | A toddler is walking near his wheelchair bound grandmother. Neither Explanation: the woman may not be his grandmother. |
| SYNTH | ADVHOTPOT | E-SNLI | ||
| OPT (175B) | FEW-SHOT | 40.52.8 | 49.72.6 | 44.03.8 |
| E-P | 29.60.5 | 52.66.5 | 39.37.8 | |
| P-E | 40.22.6 | 43.34.5 | 43.41.6 | |
| GPT-3 | FEW-SHOT | 49.50.6 | 49.16.2 | 43.35.7 |
| E-P | 47.12.8 | 54.14.1 | 40.44.5 | |
| P-E | 51.31.8 | 48.74.6 | 48.72.4 | |
| InstructGPT | FEW-SHOT | 54.83.1 | 53.22.3 | 56.82.0 |
| E-P | 58.52.1 | 58.24.1 | 41.82.5 | |
| P-E | 53.61.0 | 51.52.4 | 59.41.0 | |
| text-davinci-002 | FEW-SHOT | 72.01.4 | 77.73.2 | 69.12.0 |
| E-P | 86.93.8 | 82.45.1 | 75.67.6 | |
| P-E | 81.12.8 | 77.24.8 | 69.45.0 |
| Acc | Fac | Con | Acc=Fac | Acc=Con | ||
| InstructGPT | reliability of explanations generated by InstructGPT | |||||
| SYNTH (E-P) | 58.4 | 72.8 | 64.8 | 66.5 | 68.8 | |
| SYNTH (P-E) | 54.8 | 51.6 | 95.2 | 89.6 | 57.2 | |
| ADVHP (E-P) | 62.0 | 79.6 | 91.2 | 80.0 | 68.4 | |
| ADVHP (P-E) | 54.0 | 69.2 | 82.0 | 77.6 | 67.2 | |
| E-SNLI(P-E) | 62.0 | 1 | 98.8 | 1 | 62.0 | |
| reliability of explanations generated by other LLMs on SYNTH | ||||||
| OPT (175B) | SYNTH (E-P) | 30.0 | 77.2 | 47.2 | 45.6 | 58.8 |
| SYNTH (P-E) | 39.6 | 64.0 | 81.2 | 69.2 | 49.6 | |
| GPT-3 | SYNTH (E-P) | 46.8 | 59.2 | 64.8 | 66.8 | 61.2 |
| SYNTH (P-E) | 52.4 | 52.4 | 83.2 | 78.4 | 58.0 | |
| text-davinci-002 | SYNTH (E-P) | 86.0 | 91.6 | 85.2 | 91.2 | 84.8 |
| SYNTH (P-E) | 81.6 | 83.2 | 96.4 | 95.8 | 82.8 |
| naetatr | Pedro Rubens! The individual chapters were published into 64 "tankobon"by Kodansha. Yoko Shoji(born4June195o,in Mobara,Chiba)isaJapanese mangaartist.She is bestknown for writing "Seito Shokun! Mulder Scully!The individual chapters were published into 14 "tankobon" by Kodansha. Seito Shokun! The individual chapters were published into 24 "tankobon" by Kodansha between. Q:How many chapters does Yoko Shoji's most famous manga have? |
| Trrnsrtetetr | A: First,Yoko Shoji's mostfamous manga is "Seito Shokun!".Second,"Seito Shokun!"has 64chapters.The answer is 64. Tim Minchin (December 29,1808 July 31,1875) was the President of the United States. Andrew Johnson (December 29,1808 July 31,1875) was the President of the United States. George Andrew Atzerodt (June 12,1835-July 7,1865)was a conspirator, with John Wilkes Booth. Jesse Andrew Williams (June 12,1835-July7,1865) was a conspirator, with John Wilkes Booth. Q: Who was older, George Atzerodt or Andrew Johnson? A: First,George Atzerodt was bornonJune12,1835.Second,Andrew Johnson was bornon December 29,1808.The answer is George Atzerodt. |
| w/o Explanation | 32L | 64L | 96L | 128L |
| RoBERTa | 40.14.7 | 43.05.1 | 49.05.2 | 54.94.8 |
| FEW-SHOT | 56.82.0 | 1 | 1 | 1 |
| FEW-SHOT(NN) | 1 | 1 | 1 | 58.91.0 |
| FEW-SHOT+PROBCAL | 61.93.8 | 62.42.6 | 63.22.9 | 63.91.2 |
| w/ Explanation | 32L+32E | 64L+32E | 96L+32E | 128L+32E |
| P-E | 59.42.0 | 一 | 1 | |
| P-E+PROBCAL | 64.41.8 | 65.41.2 | 65.41.6 | 65.41.9 |
| P-E+EXPLCAL | 64.22.6 | 65.81.3 | 67.61.6 | 68.51.2 |
| P-E+ZHANG | 63.03.2 | 65.22.2 | 65.41.5 | 65.92.5 |
| w/o Explanation | 6L | 32L | 64L |
| FEW-SHOT FEW-SHOT(NN) | 59.62.4 | 1 | 1 |
| w/Explanation | 1 6L+6E | 1 32L+6E | 61.30.9 64L+6E |
| E-P | 64.42.9 | ||
| E-P+EXPLCAL | 66.03.9 | 68.83.0 | |
| E-P+ZHANG | 1 1 | 65.63.9 | 66.13.2 |
| [Factual Prompt] One of lung cancer's symptoms is shortness of breath. |
| [Greedy] The American Cancer Society estimates that more than 228,OO0 people willbe diagnosed with lung cancer in the United States this year\\n\\n The-American Cancer Society estimates-that more than 228,000 people will be diagnosed with lung cancer in the United States this year\\n\\nThe American Cancer Society estimates that more than 228,0oO people will be diagnosed with lung cancer in the United.. [p=0.9] That would make an oxygen mask one of the more popular treatments for this devastating disease. |
| It helps ease breathing and give patients back their strength. Learn more here.<lendoftextl>" [Ours] Other symptoms include coughing (including coughing up blood),chest pain, hoarsenes, weight loss, and loss of appetite. The symptoms of lung cancer may resemble those of a bad cold or pneumonia. A |
| chest x - ray is necessary to confirm the diagnosis. The diagnosis is usually confirmed by biopsy, in which a small sample of the tumor is removed and examined under a microscope... [Non-Factual Prompt] Cancer can be prevented by drinking too much alcohol. |
| [Greedy] \\nCancer can be prevented by eating too much red meat.\\n\\nCancer can be prevented by eating t00 much sugar.\\n\\nCancer can be prevented by-eating-too much fat.lnnCancer can be prevented by eating too much processed food.An\\nCaneer can be.. [p=0.9] More than 90% of hepatocellular carcinoma is associated with alcohol use.\\n\\nAbout 9 percent |
| of cancers can be prevented if we don't eat more than 1.5 ounces of processed meat each day.\\n\\nDiet is believed to play a role in 7% of cancers and... [Ours] Cancer is often treated with some combination of radiation therapy, surgery, chemotherapy and targeted therapy. Pain and symptom management are an important part of care. Palliative care is particularly |
| [Factual Prompt] One of lung cancer's symptoms is shortness of breath. |
| [Greedy] The American Cancer Society estimates that more than 228,OO0 people willbe diagnosed with lung cancer in the United States this year\\n\\n The-American Cancer Society estimates-that more than 228,000 people will be diagnosed with lung cancer in the United States this year\\n\\nThe American Cancer Society estimates that more than 228,0oO people will be diagnosed with lung cancer in the United.. [p=0.9] That would make an oxygen mask one of the more popular treatments for this devastating disease. |
| It helps ease breathing and give patients back their strength. Learn more here.<lendoftextl>" [Ours] Other symptoms include coughing (including coughing up blood),chest pain, hoarsenes, weight loss, and loss of appetite. The symptoms of lung cancer may resemble those of a bad cold or pneumonia. A |
| chest x - ray is necessary to confirm the diagnosis. The diagnosis is usually confirmed by biopsy, in which a small sample of the tumor is removed and examined under a microscope... [Non-Factual Prompt] Cancer can be prevented by drinking too much alcohol. |
| [Greedy] \\nCancer can be prevented by eating too much red meat.\\n\\nCancer can be prevented by eating t00 much sugar.\\n\\nCancer can be prevented by-eating-too much fat.lnnCancer can be prevented by eating too much processed food.An\\nCaneer can be.. [p=0.9] More than 90% of hepatocellular carcinoma is associated with alcohol use.\\n\\nAbout 9 percent |
| of cancers can be prevented if we don't eat more than 1.5 ounces of processed meat each day.\\n\\nDiet is believed to play a role in 7% of cancers and... [Ours] Cancer is often treated with some combination of radiation therapy, surgery, chemotherapy and targeted therapy. Pain and symptom management are an important part of care. Palliative care is particularly |
| Annotation | Entailr | NEER |
| Expert | 0.81 | -0.77 |
| Majority-voting | 0.47 | -0.46 |
| Annotation | Entailr | NEER |
| Expert | 0.81 | -0.77 |
| Majority-voting | 0.47 | -0.46 |
| Size | Decode | Factual Prompt | Nonfactual Prompt | ||||||
| NEER↓ | Entailr↑ | Div.↑ | Rep. | NEER | EntailR↑ | Div.个 | Rep.↓ | ||
| 126M | p=0.9 | 63.69% | 0.94% | 0.90 | 0.58% | 67.71% | 0.76% | 0.90 | 0.38% |
| greedy | 48.55% | 8.36% | 0.03 | 59.06% | 54.24% | 6.25% | 0.03 | 59.90% | |
| 357M | p=0.9 | 56.70% | 2.01% | 0.87 | 0.55% | 60.80% | 1.42% | 0.88 | 0.35% |
| greedy | 43.04% | 14.25% | 0.03 | 45.18% | 46.79% | 9.89% | 0.04 | 46.30% | |
| 1.3B | p=0.9 greedy | 52.42% | 2.93% | 0.88 | 0.24% | 56.82% | 2.04% | 0.89 | 0.25% |
| 39.87% | 12.91% | 0.05 | 33.13% | 45.02% | 8.75% | 0.05 | 36.20% | ||
| 8.3B | p=0.9 | 40.59% | 7.07% | 0.90 | 0.11% | 47.49% | 3.57% | 0.91 | 0.08% |
| greedy | 28.06% | 22.80% | 0.07 | 19.41% | 32.29% | 15.01% | 0.07 | 13.26% | |
| 530B | p=0.9 | 33.30% | 11.80% | 0.90 | 0.13% | 40.49% | 7.25% | 0.92 | 0.08% |
| greedy | 20.85% | 31.94% | 0.08 | 15.88% | 27.95% | 19.91% | 0.08 | 16.28% |
| Size | Decode | Factual Prompt | Nonfactual Prompt | ||||||
| NEER↓ | Entailr↑ | Div.↑ | Rep. | NEER | EntailR↑ | Div.个 | Rep.↓ | ||
| 126M | p=0.9 | 63.69% | 0.94% | 0.90 | 0.58% | 67.71% | 0.76% | 0.90 | 0.38% |
| greedy | 48.55% | 8.36% | 0.03 | 59.06% | 54.24% | 6.25% | 0.03 | 59.90% | |
| 357M | p=0.9 | 56.70% | 2.01% | 0.87 | 0.55% | 60.80% | 1.42% | 0.88 | 0.35% |
| greedy | 43.04% | 14.25% | 0.03 | 45.18% | 46.79% | 9.89% | 0.04 | 46.30% | |
| 1.3B | p=0.9 greedy | 52.42% | 2.93% | 0.88 | 0.24% | 56.82% | 2.04% | 0.89 | 0.25% |
| 39.87% | 12.91% | 0.05 | 33.13% | 45.02% | 8.75% | 0.05 | 36.20% | ||
| 8.3B | p=0.9 | 40.59% | 7.07% | 0.90 | 0.11% | 47.49% | 3.57% | 0.91 | 0.08% |
| greedy | 28.06% | 22.80% | 0.07 | 19.41% | 32.29% | 15.01% | 0.07 | 13.26% | |
| 530B | p=0.9 | 33.30% | 11.80% | 0.90 | 0.13% | 40.49% | 7.25% | 0.92 | 0.08% |
| greedy | 20.85% | 31.94% | 0.08 | 15.88% | 27.95% | 19.91% | 0.08 | 16.28% |
| Decoding | Factual Prompt | Nonfactual Prompt | ||||||
| NEER↓ | EntailR↑ | Div.个 | Rep. | NEER↓ | Entailr↑ | Div.↑ | Rep.↓ | |
| Greedy | 39.9% | 12.9% | 0.05 | 33.1% | 45.0% | 8.8% | 0.05 | 36.2% |
| Top-p 0.9 | 52.4% | 2.9% | 0.88 | 0.2% | 56.8% | 2.0% | 0.89 | 0.3% |
| p1 | Top-p + X-decay | |||||||
| 0.910.9 | 41.1% | 10.8% | 0.43 | 30.7% | 45.7% | 6.8% | 0.47 | 34.5% |
| 0.910.5 | 39.9% | 13.0% | 0.08 | 33.1% | 44.9% | 9.1% | 0.09 | 35.9% |
| p1入 | Top-p + X-decay + p-reset | |||||||
| 0.910.9 | 41.5% | 10.3% | 0.52 | 10.3% | 45.4% | 6.3% | 0.57 | 9.1% |
| 0.910.5 | 39.3% | 12.8% | 0.34 | 17.8% | 44.5% | 8.4% | 0.45 | 18.9% |
| pl入lw | (factual-nucleus sampling) | |||||||
| Top-p + X-decay + p-reset + w-bound | ||||||||
| 0.910.910.7 | 46.2% | 5.0% | 0.78 | 1.2% | 52.2% | 3.2% | 0.80 | 0.5% |
| 0.9 10.9 10.3 | 42.1% | 10.1% | 0.55 | 7.1% | 46.5% | 5.6% | 0.59 | 6.4% |
| 0.910.9 10.2 | 41.7% | 9.9% | 0.52 | 8.6% | 45.6% | 6.2% | 0.56 | 7.6% |
| 0.9 10.510.3 | 41.0% | 12.2% | 0.47 | 13.0% | 46.0% | 7.0% | 0.51 | 12.7% |
| 0.910.510.2 | 39.3% | 12.8% | 0.38 | 16.1% | 45.2% | 7.8% | 0.42 | 16.9% |
| Decoding | Factual Prompt | Nonfactual Prompt | ||||||
| NEER↓ | EntailR↑ | Div.个 | Rep. | NEER↓ | Entailr↑ | Div.↑ | Rep.↓ | |
| Greedy | 39.9% | 12.9% | 0.05 | 33.1% | 45.0% | 8.8% | 0.05 | 36.2% |
| Top-p 0.9 | 52.4% | 2.9% | 0.88 | 0.2% | 56.8% | 2.0% | 0.89 | 0.3% |
| p1 | Top-p + X-decay | |||||||
| 0.910.9 | 41.1% | 10.8% | 0.43 | 30.7% | 45.7% | 6.8% | 0.47 | 34.5% |
| 0.910.5 | 39.9% | 13.0% | 0.08 | 33.1% | 44.9% | 9.1% | 0.09 | 35.9% |
| p1入 | Top-p + X-decay + p-reset | |||||||
| 0.910.9 | 41.5% | 10.3% | 0.52 | 10.3% | 45.4% | 6.3% | 0.57 | 9.1% |
| 0.910.5 | 39.3% | 12.8% | 0.34 | 17.8% | 44.5% | 8.4% | 0.45 | 18.9% |
| pl入lw | (factual-nucleus sampling) | |||||||
| Top-p + X-decay + p-reset + w-bound | ||||||||
| 0.910.910.7 | 46.2% | 5.0% | 0.78 | 1.2% | 52.2% | 3.2% | 0.80 | 0.5% |
| 0.9 10.9 10.3 | 42.1% | 10.1% | 0.55 | 7.1% | 46.5% | 5.6% | 0.59 | 6.4% |
| 0.910.9 10.2 | 41.7% | 9.9% | 0.52 | 8.6% | 45.6% | 6.2% | 0.56 | 7.6% |
| 0.9 10.510.3 | 41.0% | 12.2% | 0.47 | 13.0% | 46.0% | 7.0% | 0.51 | 12.7% |
| 0.910.510.2 | 39.3% | 12.8% | 0.38 | 16.1% | 45.2% | 7.8% | 0.42 | 16.9% |
| Decoding (plλlw) | Factual Prompt | Nonfactual Prompt | ||||||
| NEER↓ | EntailR↑ | Div. | Rep. | NEER | EntailR | Div. | Rep. | |
| Vanilla Pretrained LM (1.3B) | ||||||||
| 0.9 | 52.4% | 2.9% | 0.88 | 0.2% | 56.8% | 2.0% | 0.89 | 0.3% |
| 0.9 10.910.3 | 42.1% | 10.1% | 0.55 | 7.1% | 46.5% | 5.6% | 0.59 | 6.4% |
| Factual Domain-Adaptive Training with Wikipedia (1.3B) | ||||||||
| 0.9 | 52.5% | 2.8% | 0.85 | 0.2% | 55.8% | 2.2% | 0.86 | 0.1% |
| 0.910.9 10.3 | 42.7% | 7.1% | 0.51 | 7.2% | 48.2% | 4.9% | 0.56 | 6.0% |
| TOPICPREFIX (1.3B) | ||||||||
| 0.9 0.910.910.3 | 34.4% | 4.2% | 0.84 | 0.3% | 36.2% | 2.7% | 0.85 | 0.2% |
| 27.6% | 8.7% | 0.43 | 8.0% | 30.5% | 6.1% | 0.47 | 6.9% | |
| TOPICPREFIX + SCROOT (1.3B) | ||||||||
| 0.9 0.9 10.9 10.3 | 32.5% 24.7% | 6.7% 15.8% | 0.83 0.40 | 1.2% 13.6% | 34.3% 27.6% | 4.6% 9.1% | 0.84 0.44 | 1.1% 13.7% |
| TOPICPREFIX+ SCRANDOM (1.3B) | ||||||||
| 0.9 0.9 10.9 10.3 | 32.0% 23.6% | 7.9% 17.6% | 0.81 0.39 | 1.2% 14.2% | 34.2% 26.9% | 5.5% 9.3% | 0.83 0.42 | 1.1% 13.2% |
| TOPICPREFIX + SCHALF | ||||||||
| 0.9 | 31.6% | 7.6% 17.4% | 0.81 0.38 | 1.4% 14.4% | 33.5% 27.2% | 5.1% 10.2% | 0.83 0.42 | 1.5% 13.1% |
| 0.9 10.9 10.3 23.6% Vanilla Pretrained LM (530B) | ||||||||
| 0.9 | 33.3% | 11.8% | 0.90 | 0.1% | 40.5% | 7.25% | 0.92 | 0.1% |
| TOPICPREFIX + SCHALF (530B) | ||||||||
| 0.9 0.9 10.9 10.3 | 18.3% 14.5% | 19.3% 25.5% | 0.68 0.33 | 0.1% 0.2% | 21.7% 17.7% | 13.7% 20.0% | 0.68 | 0.1% |
| Decoding (plλlw) | Factual Prompt | Nonfactual Prompt | ||||||
| NEER↓ | EntailR↑ | Div. | Rep. | NEER | EntailR | Div. | Rep. | |
| Vanilla Pretrained LM (1.3B) | ||||||||
| 0.9 | 52.4% | 2.9% | 0.88 | 0.2% | 56.8% | 2.0% | 0.89 | 0.3% |
| 0.9 10.910.3 | 42.1% | 10.1% | 0.55 | 7.1% | 46.5% | 5.6% | 0.59 | 6.4% |
| Factual Domain-Adaptive Training with Wikipedia (1.3B) | ||||||||
| 0.9 | 52.5% | 2.8% | 0.85 | 0.2% | 55.8% | 2.2% | 0.86 | 0.1% |
| 0.910.9 10.3 | 42.7% | 7.1% | 0.51 | 7.2% | 48.2% | 4.9% | 0.56 | 6.0% |
| TOPICPREFIX (1.3B) | ||||||||
| 0.9 0.910.910.3 | 34.4% | 4.2% | 0.84 | 0.3% | 36.2% | 2.7% | 0.85 | 0.2% |
| 27.6% | 8.7% | 0.43 | 8.0% | 30.5% | 6.1% | 0.47 | 6.9% | |
| TOPICPREFIX + SCROOT (1.3B) | ||||||||
| 0.9 0.9 10.9 10.3 | 32.5% 24.7% | 6.7% 15.8% | 0.83 0.40 | 1.2% 13.6% | 34.3% 27.6% | 4.6% 9.1% | 0.84 0.44 | 1.1% 13.7% |
| TOPICPREFIX+ SCRANDOM (1.3B) | ||||||||
| 0.9 0.9 10.9 10.3 | 32.0% 23.6% | 7.9% 17.6% | 0.81 0.39 | 1.2% 14.2% | 34.2% 26.9% | 5.5% 9.3% | 0.83 0.42 | 1.1% 13.2% |
| TOPICPREFIX + SCHALF | ||||||||
| 0.9 | 31.6% | 7.6% 17.4% | 0.81 0.38 | 1.4% 14.4% | 33.5% 27.2% | 5.1% 10.2% | 0.83 0.42 | 1.5% 13.1% |
| 0.9 10.9 10.3 23.6% Vanilla Pretrained LM (530B) | ||||||||
| 0.9 | 33.3% | 11.8% | 0.90 | 0.1% | 40.5% | 7.25% | 0.92 | 0.1% |
| TOPICPREFIX + SCHALF (530B) | ||||||||
| 0.9 0.9 10.9 10.3 | 18.3% 14.5% | 19.3% 25.5% | 0.68 0.33 | 0.1% 0.2% | 21.7% 17.7% | 13.7% 20.0% | 0.68 | 0.1% |
| Methods | Structure | Attention Types | Windows of Local Attention | Handling Non-overlapped Regions | Global Tokens |
| ViT (Dosovitskiy et al.,2021) | Isotropic | Global | N/A | N/A | N/A |
| PVT (Wang et al., 2021) | Pyramid | Global | N/A | N/A | N/A |
| Swin (Liu et al.,2021) | Pyramid | Local | Strictly Non-overlapped | Shifting the window | None |
| ViL (Zhang et al., 2021a) | Pyramid | Local + Global | Overlapped | N/A1 | One learnable token |
| Twins (Chu et al., 2021a) | Pyramid | Local + Global | Non-overlapped | GA | Subsampled from local tokens |
| RegionViT (Ours) | Pyramid | Local + Regional | Non-overlapped | Regional-to-local attention | Regional tokens |
| Model | Tokenization Local Regional | Window size (M) | Dimension (C) | # of Encoders at each stage | |
| RegionViT-Ti | 3-conv | linear | 7 | {64,128,256,512} | {2,2,8,2} |
| RegionViT-S | 3-conv | linear | 7 | {96,192,384,768} | {2,2,8,2} |
| RegionViT-M | 1-conv | linear | 7 | {96,192,384,768} | {2,2,14,2} |
| RegionViT-B | 1-conv | linear | 7 | {128,256,512,1024} | {2,2,14,2} |
| Model | Params (M) | FLOPs (G) | Acc. (%) |
| PiT-XS (Heo et al., 2021) | 10.6 | 1.4 | 78.1 |
| ConT-S(Yan et al.,2021) | 10.1 | 1.5 | 76.5 |
| PVT-T(Wang et al.,2021) | 13.2 | 1.9 | 75.1 |
| ConViT-Ti+ (d'Ascoli et al., 2021) | 10.0 | 2.0 | 76.7 |
| Twins-SVT-S (Chu et al.,2021a) | 24.0 | 2.8 | 81.7 |
| PiT-S (Heo et al., 2021) | 23.5 | 2.9 | 80.9 |
| ConT-M(Yan et al., 2021) | 19.2 | 3.1 | 80.2 |
| Twins-PCPVT-S (Chu et al., 2021a) | 24.1 | 3.7 | 81.2 |
| PVT-S (Wang et al., 2021) | 24.5 | 3.8 | 79.8 |
| RegionViT-Ti | 13.8 | 2.4 | 80.4 |
| RegionViT-Ti+ | 14.3 | 2.7 | 81.5 |
| DeiT-S (Touvron et al., 2020) | 22.1 | 4.6 | 79.9 |
| CvT-13 (Wu et al.,2021) | 20.0 | 4.5 | 81.6 |
| Swin-T (Liu et al., 2021) | 29.0 | 4.5 | 81.3 |
| LocalViT-S (Li et al.,2021) | 22.4 | 4.6 | 80.8 |
| ViL-S (Zhang et al., 2021a) | 24.6 | 4.9 | 82.0 |
| Visformer-S (Chen et al.,2021b) | 40.2 | 4.9 | 82.3 |
| ConViT-S (d'Ascoli et al.,2021) | 27.0 | 5.4 | 81.3 |
| NesT-S (Zhang et al., 2021b) | 17.0 | 5.8 | 81.5 |
| RegionViT-S | 30.6 | 5.3 | 82.6 |
| RegionViT-S+ | 31.3 | 5.7 | 83.3 |
| Model | Params (M) | FLOPs (G) | Acc. (%) |
| ConT-M(Yan et al., 2021) | 39.6 | 6.4 | 81.8 |
| Twins-PCPVT-B (Chu et al.,2021a) | 43.8 | 6.4 | 82.7 |
| PVT-M(Wang et al., 2021) | 44.2 | 6.7 | 81.2 |
| CvT-21(Wu et al.,2021) | 32.0 | 7.1 | 82.5 |
| Twins-SVT-B (Chu et al., 2021a) | 56 | 8.3 | 83.2 |
| ViL-M (Zhang et al.,2021a) | 39.7 | 8.7 | 83.3 |
| Swin-S (Liu et al., 2021) | 50.0 | 8.7 | 83.0 |
| PVT-L (Wang et al.,2021) | 61.4 | 9.8 | 81.7 |
| ConViT-S+ (d'Ascoli et al., 2021) | 48.0 | 10.0 | 82.2 |
| NesT-S (Zhang et al., 2021b) | 38.0 | 10.4 | 83.3 |
| RegionViT-M | 41.2 | 7.4 | 83.1 |
| RegionViT-M+ | 42.0 | 7.9 | 83.4 |
| DeiT-B (Touvron et al.,2020) | 86.6 | 17.6 | 81.8 |
| PiT-B (Heo et al., 2021) | 73.8 | 12.5 | 82.0 |
| ViL-B (Zhang et al., 2021a) | 55.7 | 13.4 | 83.2 |
| Twins-SVT-L(Chu et al.,2021a) | 99.2 | 14.8 | 83.7 |
| Swin-B (Liu et al.,2021) | 88.0 | 15.4 | 83.5 |
| ConViT-B (d'Ascoli et al.,2021) | 86.0 | 17.0 | 82.4 |
| NesT-B (Zhang et al., 2021b) | 68.0 | 17.9 | 83.8 |
| RegionViT-B | 72.7 | 13.0 | 83.2 |
| RegionViT-B+ | 73.8 | 13.6 | 83.8 |
| Model | Params (M) | FLOPs (G) | Acc. (%) |
| ViL-B | 55.7 | 43.7 | 86.0 |
| Swin-L | 197.0 | 103.9 | 87.3 |
| CvT-W24 | 277 | 193.2 | 87.7 |
| RegionViT-B+ | 76.5 | 42.6 | 86.5 |
| CIFAR10 | CIFAR100 | Pet | StanfordCars | ChestXRay8 | |
| DeiT-S | 99.1 | 90.9 | 94.9 | 91.5 | 55.4 |
| DeiT-B | 99.1 | 90.8 | 94.4 | 91.7 | 55.8 |
| RegionViT-S | 98.9 | 90.0 | 95.3 | 92.8 | 57.8 |
| RegionViT-M | 99.0 | 90.8 | 95.5 | 91.9 | 58.3 |
| Backbone | Params | FLOPs | RetineNet | 3× AP | Params (M) | FLOPs | MaskRCNN | ||||
| (M) | 1× AP | (G) | 1× APb |APm | APb | 3× APm | ||||||
| ResNet50 | 37.7 | 234 | 36.3 | 39.0 | 44.2 | 260.0 | 38.0 | 34.4 | 41.0 | ||
| ConT-M (Yan et al.,2021) | 27.0 | 217.2 | 39.3 | 1 | 1 | 1 | 40.5 | 38.1 | 1 | 37.1 1 | |
| PVT-S (Wang et al., 2021) | 34.2 | 40.4 | 42.2 | 44.1 | 40.4 | 37.8 | 43.0 | 39.6 | |||
| ViL-S (Zhang et al., 2021a) | 35.7 | 252.2 | 41.6 | 42.9 | 45.0 | 174.3 | 41.8 | 38.5 | 43.4 | 39.6 | |
| Swin-T (Liu et al., 2021) | 38.5 | 245 | 41.5 | 43.9 | 47.8 | 264 | 42.2 | 39.1 | 46.0 | 41.6 | |
| Twins-SVT-S (w/PEG) (Chu et al.,2021a) | 34.3 | 209 | 43.0 | 45.6 | 44.0 | 228 | 43.5 | 40.3 | 46.8 | 42.6 | |
| RegionViT-S | 40.8 | 192.6 | 42.2 | 45.8 | 50.1 | 171.3 | 42.5 | 39.5 | 46.3 | 42.3 | |
| RegionViT-S+ | 41.5 | 204.2 | 43.1 | 46.9 | 50.9 | 182.9 | 43.5 | 40.4 | 47.3 | 43.4 | |
| RegionViT-S+ w/PEG | 41.6 | 204.3 | 43.9 | 46.7 | 50.9 | 183.0 | 44.2 | 40.8 | 47.6 | 43.4 | |
| ResNet101 | 56.7 | 315 | 38.5 | 40.9 | 63.2 | 336 | 40.4 | 36.4 | 42.8 | 38.5 | |
| ResNeXt101-32x4d | 56.4 | 319.1 | 39.9 | 41.4 | 62.8 | 340 | 41.9 | 37.5 | 44.0 | 39.2 | |
| PVT-M(Wang et al., 2021) | 53.9 | 41.9 | 43.2 | 63.9 | 42.0 | 39.0 | 44.2 | 40.5 | |||
| ViL-M (Zhang et al.,2021a) | 50.8 | 338.9 | 42.9 | 43.7 | 60.1 | 261.1 | 43.4 | 39.7 | 44.6 | 40.7 | |
| Swin-S (Liu et al., 2021) | 59.8 | 335 | 44.5 | 46.3 | 69.1 | 354 | 44.8 | 40.9 | 47.6 | 42.8 | |
| Twins-SVT-B (w/PEG) (Chu et al., 2021a) | 67.0 | 322 | 45.3 | 46.9 | 76.3 | 340 | 45.2 | 41.5 | 48.0 | 43.0 | |
| RegionViT-B | 83.4 | 308.9 | 43.3 | 46.1 | 92.2 | 287.9 | 43.5 | 40.1 | 47.2 | 43.0 | |
| RegionViT-B+ | 84.4 | 328.1 | 44.2 | 46.9 | 93.2 | 307.1 | 44.5 | 41.0 | 48.1 | 43.5 | |
| RegionViT-B+ w/PEG | 84.5 | 328.2 | 44.6 | 46.9 | 93.2 | 307.2 | 45.4 | 41.6 | 48.3 | 43.5 | |
| ResNeXt101-64x4d | 95.5 | 473 | 41.0 | 1 | 101.9 | 493 | 42.8 | 38.4 | |||
| PVT-L (Wang et al., 2021) | 71.1 | 345 | 42.6 | 81.0 | 364 | 42.9 | 39.5 | 1 | 1 | ||
| ViL-B (Zhang et al., 2021a) | 66.7 | 443.0 | 44.3 | 44.7 | 76.1 | 365.1 | 45.1 | 41.0 | 45.7 | 41.3 | |
| Swin-B (Liu et al.,2021) | 98.4 | 477 | 44.7 | 1 | 107.2 | 496 | 45.5 | 41.3 | 二 | ||
| Twins-SVT-L (w/ PEG) (Chu et al.,2021a) | 110.9 | 455 | 45.7 | 119.7 | 474 | 45.9 | 41.6 | 1 | |||
| RegionViT-B+ w/PEGt | 84.5 | 506.4 | 46.1 | 48.2 | 93.2 | 464.4 | 46.3 | 42.4 | 49.2 | 44.5 | |
| Model | FLOPs (G) | Params (M) | mIoU (%) |
| Swin-T* | 46 | 31.9 | 41.5 |
| Twins-SVT-S | 37 | 28.3 | 43.2 |
| RegionViT-S+ (w/PEG) | 37 | 35.7 | 45.3 |
| Swin-S* | 70 | 53.2 | 45.2 |
| Twins-SVT-B | 67 | 60.4 | 45.3 |
| RegionViT-B+ (w/PEG) | 67 | 78.3 | 47.5 |
| Model | FLOPs* (G) | Params (M) | K400 Acc. (%) | SSV2 Acc.(%) |
| TimeSformer | 197 | 121.4 | 75.8 | 59.5 |
| x TimeSformer† | 197 | 121.4 | 77.1 | 59.2 |
| RegionViT-S | 59.4 | 42.9 | 76.6 | 59.7 |
| RegionViT-M | 83.1 | 57.5 | 77.6 | 59.8 |
| Dataset Regional Tokens | IN1K Acc. | MS COCO MaskRCNN RetinaNet | ADE20K SemanticFPN | |
| (APb/APm) 40.5/37.8 | (APb) 40.7 | (mIoU) 42.3 | ||
| N Y | 82.2 82.6 | 42.5/39.5 | 42.2 | 43.7 |
| Downsampling | Params | FLOPs (G) | IN1K Acc. (%) |
| 2×2 convolution | 32.1 | 5.5 | 82.3 |
| 3×3 convolution | 34.1 | 5.7 | 82.5 |
| 3×3 depth-wise conv. | 30.6 | 5.3 | 82.6 |
| Weight sharing | Params | FLOPs (G) | IN1K Acc. (%) |
| N | 40.4 | 5.3 | 82.7 |
| Y | 30.6 | 5.3 | 82.6 |
| Regional tokens | Params | FLOPs (G) | IN1K Acc. (%) |
| All | 30.6 | 6.0 | 82.7 |
| Corresponding | 30.6 | 5.3 | 82.6 |
| Abs. pos. | Rel. pos. | Params (M) | FLOPs (G) | IN1K Acc. (%) |
| N | N | 30.6 | 5.3 | 82.4 |
| Y | N | 30.9 | 5.3 | 82.2 |
| Y | Y | 30.9 | 5.3 | 82.7 |
| N | Y | 30.6 | 5.3 | 82.6 |
| Model | Params | FLOPs (G) | IN1K Acc. (%) |
| overlapped | 30.6 | 18.8 | 82.8 |
| non-overlapped | 30.6 | 5.3 | 82.6 |
| Model | Params | FLOPs (G) | IN1K Acc. (%) |
| Shifted-window | 29.0 | 4.5 | 81.3 |
| R2L | 32.1 | 5.2 | 82.0 |
| IN1K | IN21K | Finetune to IN1K@384² | Transfer | |
| Batch size | 4,096 | 4,096 | 1,024 | 768 |
| Epochs | 300 | 120 | 30 | 1,000 |
| Optimizer | AdamW | AdamW | AdamW | SGD |
| Weight Decay | 0.05 | 0.01 | 1e-8 | 1e-4 |
| Linear-rate Scheduler (Initial LR) | Cosine (0.004) | Cosine (0.001) | Cosine (0.0002) | Cosine (0.01) |
| Warmup Epochs | 50 | 5 | 0 | 5 |
| Warmup linear-rate | Linear (1e-6) | |||
| Scheduler (Initial LR) | ||||
| Data Aug. | RandAugment (m=9,n=2) | |||
| Mixup (α) | 0.8 | |||
| CutMix (α) Random Erasing | 1.0 0.25 | |||
| Instance | ||||
| Repetition* | 3 | |||
| Drop-path | 0.1 | |||
| Label Smoothing | 0.0 0.1 | |||
| Backbone | Params (M) | FLOPs (G) | AP AP50 | RetineNet AP75 APs | Params | FLOPs | MaskRCNN AP5 | |||||||||
| 1 | APM | APL | (M) | (G) | APb | AP | APm | AP0 | APm5 | |||||||
| ResNet50 | 37.7 | 234 | 36.3 | 55.3 | 38.6 | 19.3 | 40.0 | 48.8 | 44.2 | 260.0 | 38.0 | 58.6 | 41.4 | 34.4 | 55.1 | 36.7 |
| ConT-M(Yan et al.,2021) | 27.0 | 217.2 | 39.3 | 59.3 | 41.8 | 23.1 | 43.1 | 51.9 | 40.5 | 一 | 38.1 | |||||
| PVT-S (Wang et al.,2021) | 34.2 | 40.4 | 61.3 | 43.0 | 25.0 | 42.9 | 55.7 | 44.1 | 40.4 | 62.9 | 43.8 | 37.8 | 60.1 | 40.3 | ||
| ViL-S (Zhang et al.,2021a) | 35.7 | 252.2 | 41.6 | 62.5 | 44.1 | 24.9 | 44.6 | 56.2 | 45.0 | 174.3 | 41.8 | 64.1 | 45.1 | 38.5 | 61.1 | 41.4 |
| Swin-T(Liu et al.,2021) | 38.5 | 245 | 41.5 | 62.1 | 44.2 | 25.1 | 44.9 | 55.5 | 47.8 | 264 | 42.2 | 64.6 | 46.2 | 39.1 | 61.6 | 42.0 |
| Twins-SVT-S(w/PEG)(Chu etal.,21a) | 34.3 | 209 | 43.0 | 64.2 | 46.3 | 28.0 | 46.4 | 57.5 | 44.0 | 228 | 43.5 | 66.0 | 47.3 | 40.3 | 63.2 | 43.4 |
| RegionViT-S | 40.8 | 192.6 | 42.2 | 64.1 | 45.1 | 27.5 | 45.4 | 55.3 | 50.1 | 171.3 | 42.5 | 65.8 | 46.1 | 39.5 | 62.8 | 42.2 |
| RegionViT-S+ | 41.5 | 204.2 | 43.1 | 64.8 | 46.2 | 29.6 | 46.6 | 56.1 | 50.9 | 182.9 | 43.5 | 66.9 | 47.5 | 40.4 | 63.7 | 43.4 |
| RegionViT-S+W/PEG | 41.6 | 204.3 | 43.9 | 65.5 | 47.3 | 28.5 | 47.3 | 57.9 | 50.9 | 183.0 | 44.2 | 67.3 | 48.2 | 40.8 | 64.1 | 44.0 |
| ResNet101 | 315 | 57.7 | 38.8 | |||||||||||||
| ResNeXt101-32x4d | 56.7 56.4 | 319.1 | 38.5 39.9 | 57.8 59.6 | 41.2 42.7 | 21.4 22.3 | 42.6 44.2 | 51.1 52.5 | 63.2 | 336 | 40.4 41.9 | 61.1 62.5 | 44.2 | 36.4 37.5 | 59.4 | 40.2 |
| PVT-M(Wang et al.,2021) | 53.9 | 41.9 | 63.1 | 44.3 | 25.0 | 44.9 | 57.6 | 62.8 63.9 | 340 | 42.0 | 64.4 | 45.9 45.6 | 39.0 | 61.6 | 42.1 | |
| ViL-M(Zhang et al.,021a) | 50.8 | 338.9 | 42.9 | 64.0 | 45.4 | 27.0 | 46.1 | 57.2 | 60.1 | 261.1 | 43.4 | 65.9 | 47.0 | 39.7 | 62.8 | 42.1 |
| Swin-S (Liu et al.,2021) | 59.8 | 335 | 44.5 | 65.7 | 47.5 | 27.4 | 48.0 | 59.9 | 69.1 | 354 | 44.8 | 66.6 | 48.9 | 40.9 | 63.4 | 44.2 |
| Twins-SVT-B(w/PEG)(Chu etal.,221a) | 67.0 | 322 | 45.3 | 66.7 | 48.1 | 28.5 | 48.9 | 60.6 | 76.3 | 340 | 45.2 | 67.6 | 49.3 | 41.5 | 64.5 | 44.8 |
| RegionViT-B | 83.4 | 308.9 | 43.3 | 65.2 | 46.4 | 29.2 | 46.4 | 57.0 | 92.2 | 287.9 | 43.5 | 66.7 | 47.4 | 40.1 | 63.4 | 43.0 |
| RegionViT-B+ | 84.4 | 328.1 | 44.2 | 66.2 | 47.1 | 29.2 | 47.5 | 58.6 | 93.2 | 307.1 | 44.5 | 67.6 | 48.7 | 41.0 | 64.4 | 43.9 |
| RegionViT-B+ W/PEG | 84.5 | 328.2 | 44.6 | 66.4 | 47.6 | 29.6 | 47.6 | 59.0 | 93.2 | 307.2 | 45.4 | 68.4 | 49.6 | 41.6 | 65.2 | 44.8 |
| ResNeXt101-64x4d | 95.5 | 473 | 41.0 | 60.9 | 44.0 | 23.9 | 45.2 | 54.0 | 101.9 | 493 | 42.8 | 63.8 | 47.3 | 38.4 | 60.6 | 41.3 |
| PVT-L (Wang et al., 2021) | 71.1 | 345 | 42.6 | 63.7 | 45.4 | 25.8 | 46.0 | 58.4 | 81.0 | 364 | 42.9 | 65.0 | 46.6 | 39.5 | 61.9 | 42.5 |
| ViL-B (Zhang et al., 2021a) | 66.7 | 443.0 | 44.3 | 65.5 | 47.1 | 28.9 | 47.9 | 58.3 | 76.1 | 365.1 | 45.1 | 67.2 | 49.3 | 41.0 | 64.3 | 44.2 |
| Swin-B (Liu etal.,2021) | 98.4 | 477 | 44.7 | 65.9 | 49.2 | 1 | 1 | 1 | 107.2 | 496 | 45.5 | 1 | 1 | 41.3 | 1 | 1 |
| Twins-SVT-L (w/PEG) (Chu et al.,2021a) | 110.9 | 455 | 45.7 | 67.1 | 49.2 | 一 | 一 | 一 | 119.7 | 474 | 45.9 | 一 | 一 | 41.6 42.4 | 66.2 | 45.6 |
| RegionViT-B+W/PEGt | 84.5 | 506.4 | 46.1 | 68.0 | 49.5 | 30.5 | 49.9 | 60.1 | 93.2 | 464.4 | 46.3 | 69.1 | 51.2 | |||
| Backbone | Params (M) | FLOPs (G) | AP AP50 | RetineNet AP75|APs | Params (M) | FLOPs G | APb | MaskRCNN | |||||||||
| APM| | APL | AP | AP5 | |APm | AP | APm5 | |||||||||||
| ResNet50 | 37.7 | 234 | 39.0 | 58.4 | 41.8 | 22.4 | 42.8 | 51.6 | 44.2 | 260.0 | 41.0 | 61.7 | 44.9 | 37.1 | 58.4 | 40.1 | |
| PVT-S (Wang etal.,2021) | 34.2 | 1 | 42.2 | 62.7 | 45.0 | 26.2 | 45.2 | 57.2 | 44.1 | 245 | 43.0 | 65.3 | 46.9 | 39.9 | 62.5 | 42.8 | |
| ViL-S (Zhang etal.,2021a) | 35.7 | 252.2 | 42.9 | 63.8 | 45.6 | 27.8 | 46.4 | 56.3 | 45.0 | 174.3 | 43.4 | 64.9 | 47.0 | 39.6 | 62.1 | 42.4 | |
| Swin-T(Liu et al., 2021) | 38.5 | 245 | 43.9 | 64.8 | 47.1 | 28.4 | 47.2 | 57.8 | 47.8 | 264 | 46.0 | 68.2 | 50.2 | 41.6 | 65.1 | 44.8 | |
| Twins-SVT-S (w/PEG)(Chu et al., 2021a) | 34.3 | 209 | 45.6 | 67.1 | 48.6 | 29.8 | 49.3 | 60.0 | 44.0 | 228 | 46.8 | 69.2 | 51.2 | 42.6 | 66.3 | 45.8 | |
| RegionViT-S | 40.8 | 192.6 | 45.8 | 67.2 | 49.2 | 30.0 | 50.0 | 60.5 | 50.1 | 171.3 | 46.3 | 68.8 | 50.6 | 42.3 | 65.5 | 45.7 | |
| RegionViT-S+ | 41.5 | 204.2 | 46.9 | 68.3 | 50.7 | 31.1 | 51.0 | 62.0 | 50.9 | 182.9 | 47.3 | 69.5 | 52.0 | 43.1 | 66.4 | 46.7 | |
| RegionViT-S+ w/PEG | 41.6 | 204.3 | 46.7 | 68.2 | 50.2 | 30.7 | 50.8 | 62.4 | 50.9 | 183.0 | 47.6 | 70.0 | 52.0 | 43.4 | 67.1 | 47.0 | |
| ResNet101 | 56.7 | 315 | 40.9 | 44.0 | 63.2 | 336 | 42.8 | 47.1 | 38.5 | 60.1 | 41.3 | ||||||
| ResNeXt101-32x4d | 56.4 | 319.1 | 41.4 | 60.1 61.0 | 44.3 | 23.7 23.9 | 45.0 45.5 | 53.8 53.7 | 62.8 | 340 | 44.0 | 63.2 64.4 | 48.0 | 39.2 | 61.4 | 41.9 | |
| PVT-M(Wang etal.,021) | 53.9 | 1 | 43.2 | 63.8 | 46.1 | 27.3 | 46.3 | 58.9 | 63.9 | 302 | 44.2 | 66.0 | 48.2 | 40.5 | 63.1 | 43.5 | |
| ViL-M(Zhang etal.,2021a) | 50.8 | 338.9 | 43.7 | 64.6 | 46.4 | 27.9 | 47.1 | 56.9 | 60.1 | 261.1 | 44.6 | 66.3 | 48.5 | 40.7 | 63.8 | 43.7 | |
| Swin-S (Liuetal.,221) | 59.8 | 335 | 46.3 | 67.4 | 49.8 | 31.1 | 50.3 | 60.9 | 69.1 | 354 | 47.6 | 69.4 | 52.5 | 42.8 | 66.5 | 46.4 | |
| Twins-SVT-B (w/ PEG)(Chu et al., 2021a) | 67.0 | 322 | 46.9 | 68.0 | 50.2 | 31.7 | 50.3 | 61.8 | 76.3 | 340 | 48.0 | 69.5 | 52.7 | 43.0 | 66.8 | 46.6 | |
| RegionViT-B | 83.4 | 308.9 | 46.1 | 67.8 | 49.1 | 31.5 | 50.2 | 61.2 | 92.2 | 287.9 | 47.2 | 69.1 | 51.7 | 43.0 | 66.4 | 46.5 | |
| RegionViT-B+ | 84.4 | 328.1 | 46.9 | 68.6 | 50.1 | 30.8 | 50.7 | 62.6 | 93.2 | 307.1 | 48.1 | 70.2 | 52.5 | 43.5 | 67.1 | 47.1 | |
| RegionViT-B+ W/PEG | 84.5 | 328.2 | 46.9 | 68.3 | 50.3 | 31.1 | 50.5 | 62.4 | 93.2 | 307.2 | 48.3 | 70.1 | 52.8 | 43.5 | 67.1 | 47.0 | |
| ViL-B(Zhang etal.,2021a) | 66.7 | 443.0 | 44.7 | 65.5 | 47.6 | 29.9 | 48.0 | 58.1 | 76.1 | 365.1 | 45.7 | 67.2 | 49.9 | 41.3 | 64.4 | 44.5 | |
| RegionViT-B+w/PEGt | 84.5 | 506.4 | 48.2 | 69.9 | 51.5 | 34.3 | 51.5 | 61.7 | 93.2 | 464.4 | 49.2 | 71.0 | 53.7 | 44.5 | 68.4 | 48.3 | |
| Model | FLOPs (G) | Params (M) | bbox AP (%) | keypoint AP (%) |
| ResNet-50 | 137.7 | 59.1 | 53.6 | 64.0 |
| RegionViT-S+ (w/ PEG) | 172.2 | 65.7 | 56.0 | 66.1 |
| Regional TokensFLOPs (G) Params (M)ImageNet1K Acc.(%) | |||
| N | 5.0 | 30.4 | 82.2 |
| Y | 5.3 | 30.6 | 82.6 |
| MaskRCNN | MS COCO (APb/APm) | ||
| N | 168.5 | 49.9 | 40.5/37.8 |
| Y | 171.3 | 50.1 | 42.2/39.0 |
| RetinaNet | MS COCO (APb) | ||
| N | 189.8 | 40.5 | 40.7 |
| Y | 192.6 | 40.8 | 41.7 |
| SemanticFPN | ADE20K (mIoU) | ||
| N | 36.8 | 34.7 | 42.3 |
| Y | 37.3 | 35.0 | 43.7 |
| Model | Acc. (%) | Params (M) | FLOPs (G) | Throughput (images/sec) |
| Swin-T | 81.3 | 29.0 | 4.5 | 1129 |
| Swin-S | 83.0 | 50.0 | 8.7 | 710 |
| RegionViT-S | 82.6 | 30.6 | 5.3 | 823 |
| RegionViT-M | 83.1 | 41.2 | 7.4 | 706 |
| Year | Papers | Year | Papers |
| 2010-2016 | 11 | 2020 | 68 |
| 2017 | 16 | 2021 | 96 |
| 2018 | 10 | 2022 | 79 |
| 2019 | 46 | 2023 | 22 |
| Totals | 83 | Totals | 265 |
| Type | Papers | % |
| Measurement Debiasing ／Mitigation | 98 52 | 46.66 24.76 |
| Combinations of above | 32 | 15.23 |
| Detection | 19 | 9.04 |
| Generation | 5 | 2.38 |
| Miscellaneous | 4 | 1.9 |
| Total | 210 | 100.0 |
| prrreeade rabrrgsmrs erdnei arrreeeeeingseeenntsuorsseioidnarlgeiroiga[ergiiedJhnrittieaenasrlqesipGrariiagepueeerpeateaaeitueung2 | ||||||||||||||||||
| 19.3 | 8.2 | 8.8 | -8.5 | 5.8 | -8.6 | 3.0 | -6.5 | -7.5 | -6.5 | 5.1 | 1.7 | 3.8 | 3.0 | 1.1 | -2.3 | -1.7 | gender | |
| 19.3 | 12.2 | 12.4 | -1.0 | 6.8 | -3.0 | 4.2 | -1.6 | -2.6 | -2.6 | 8.6 | 3.8 | 4.2 | 4.6 | 0.0 | -0.8 | -0.6 | race | |
| 8.2 | 12.2 | 4.4 | -1.4 | 7.9 | -1.0 | 7.0 | -0.9 | -0.9 | -0.9 | 4.2 | 4.2 | 2.4 | 5.5 | 0.7 | -0.3 | -0.2 | religion | |
| 8.8 | 12.4 | 4.4 | -1.3 | 3.9 | -1.0 | 6.1 | 0.1 | -0.9 | 0.1 | 4.2 | 5.3 | 4.4 | 5.5 | 0.7 | -0.3 | -0.2 | age | |
| -8.5 | -1.0 | -1.4 | -1.3 | 0.1 | 0.1 | 0.2 | -0.7 | -0.7 | -0.7 | -0.7 | -0.6 | -0.5 | -0.4 | -0.3 | -0.2 | 0.8 | sentiment | |
| 5.8 | 6.8 | 7.9 | 3.9 | 0.1 | -0.7 | 5.4 | -0.6 | -0.6 | -0.6 | 3.4 | 4.5 | 0.6 | 4.7 | -0.2 | -0.2 | -0.1 | profession | |
| -8.6 | -3.0 | -1.0 | -1.0 | 0.1 | -0.7 | -0.6 | -0.6 | -0.6 | -0.6 | -0.5 | -0.5 | -0.4 | -0.3 | -0.2 | -0.2 | -0.1 | media | |
| 3.0 | 4.2 | 7.0 | 6.1 | 0.2 | 5.4 | -0.6 | -0.5 | -0.5 | -0.5 | 3.5 | 5.6 | 3.6 | 4.7 | -0.2 | -0.2 | -0.1 | nationality | |
| -6.5 | -1.6 | -0.9 | 0.1 | -0.7 | -0.6 | -0.6 | -0.5 | -0.5 | 0.5 | 0.6 | 0.6 | 0.7 | -0.3 | -0.2 | -0.1 | -0.1 | political | |
| -7.5 | -2.6 | -0.9 | -0.9 | -0.7 | -0.6 | -0.6 | -0.5 | -0.5 | -0.5 | -0.4 | -0.4 | -0.3 | -0.3 | -0.2 | -0.1 | -0.1 | multiplesocial | |
| -6.5 | -2.6 | -0.9 | 0.1 | -0.7 | -0.6 | -0.6 | -0.5 | 0.5 | -0.5 | -0.4 | -0.4 | -0.3 | -0.3 | -0.2 | -0.1 | -0.1 | annotator | |
| 5.1 | 8.6 | 4.2 | 4.2 | -0.7 | 3.4 | -0.5 | 3.5 | 0.6 | -0.4 | -0.4 | 4.6 | 0.7 | 3.8 | -0.2 | -0.1 | -0.1 | sexuality | |
| 1.7 | 3.8 | 4.2 | 5.3 | -0.6 | 4.5 | -0.5 | 5.6 | 0.6 | -0.4 | -0.4 | 4.6 | 1.7 | 4.8 | -0.2 | -0.1 | -0.1 | disability | |
| 3.8 | 4.2 | 2.4 | 4.4 | -0.5 | 0.6 | -0.4 | 3.6 | 0.7 | -0.3 | -0.3 | 0.7 | 1.7 | 0.8 | -0.1 | -0.1 | -0.1 | ethnicity | |
| 3.0 | 4.6 | 5.5 | 5.5 | -0.4 | 4.7 | -0.3 | 4.7 | -0.3 | -0.3 | -0.3 | 3.8 | 4.8 | 0.8 | -0.1 | -0.1 | -0.1 | physicalappearance | |
| 1.1 | 0.0 | 0.7 | 0.7 | -0.3 | -0.2 | -0.2 | -0.2 | -0.2 | -0.2 | -0.2 | -0.2 | -0.2 | -0.1 | -0.1 | -0.1 | -0.0 | dialect | |
| -2.3 | -0.8 | -0.3 | -0.3 | -0.2 | -0.2 | -0.2 | -0.2 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | human | |
| -1.7 | -0.6 | -0.2 | -0.2 | 0.8 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | intersectional | |
| en | de | sp | zh1 | fr | it | ru | tr | pl | nl | ar | pt | da | te | SV | no | ja | hi | sk | ko | he | ||
| 0.3 | 2.1 | 2.9 | -4.2 | 1.6 | 0.8 | -1.2 | 1.5 | -0.7 | -1.7 | -0.7 | 0.0 | -2.0 | -2.2 | -1.2 | -1.2 | -1.2 | -2.2 | -0.5 | 0.5 | 0.5 | en | |
| 0.3 | 2.7 | 3.7 | 1.8 | 1.8 | 0.8 | 0.9 | 0.9 | 1.9 | -0.1 | 2.9 | 0.9 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | 1.0 | ||
| 2.1 | 2.7 | 8.7 | 5.0 | 7.0 | 4.2 | 4.5 | 4.6 | 1.7 | 1.7 | 2.7 | 2.7 | 0.7 | -0.2 | -0.2 | -0.2 | 1.8 | -0.2 | -0.1 | 0.9 | 1.9 | de | |
| 2.9 | 3.7 | 8.7 | 3.2 | 6.2 | 5.4 | 3.6 | 3.7 | 1.7 | -0.3 | 3.7 | 3.8 | 0.8 | -0.2 | -0.2 | -0.2 | 0.8 | -0.2 | -0.1 | 0.9 | 1.9 | sp | |
| -4.2 | 1.8 | 5.0 | 3.2 | 1.4 | 0.5 | 1.7 | 1.7 | 0.8 | -0.2 | 2.8 | 1.8 | 0.8 | -0.1 | 0.9 | -0.1 | 1.9 | -0.1 | -0.1 | 0.9 | -0.1 | zh | |
| 1.6 | 1.8 | 7.0 | 6.2 | 1.4 | 6.6 | 1.7 | -0.2 | -0.2 | 0.8 | 1.8 | 0.8 | 0.8 | -0.1 | -0.1 | 0.9 | -0.1 | -0.1 | -0.1 | -0.1 | 0.9 | fr | |
| 0.8 | 0.8 | 4.2 | 5.4 | 0.5 | 6.6 | 1.8 | 0.8 | 0.8 | 0.8 | 0.8 | 1.9 | 0.9 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 -0.1 | -0.1 | 0.9 | it | |
| -1.2 | 0.9 | 4.5 | 3.6 | 1.7 | 1.7 | 1.8 | 1.9 | -0.1 | -0.1 | 1.9 | 0.9 | 0.9 | -0.1 | 0.9 | -0.1 | 0.9 | -0.1 | -0.0 | -0.0 | 1.0 | ru | |
| 1.5 | 0.9 | 4.6 | 3.7 | 1.7 | -0.2 | 0.8 | 1.9 | 1.9 | -0.1 | -0.1 | 0.9 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | 1.0 | -0.0 | tr | |
| -0.7 | 1.9 | 1.7 | 1.7 | 0.8 | -0.2 | 0.8 | -0.1 | 1.9 | -0.1 | -0.1 | 1.9 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | pl | |
| -1.7 | -0.1 | 1.7 | -0.3 | -0.2 | 0.8 | 0.8 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | nl | |
| -0.7 | 2.9 | 2.7 | 3.7 | 2.8 | 1.8 | 0.8 | 1.9 | -0.1 | -0.1 | -0.1 | 0.9 | -0.1 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | 1.0 | ar | |
| 0.0 | 0.9 | 2.7 | 3.8 | 1.8 | 0.8 | 1.9 | 0.9 | 0.9 | 1.9 | -0.1 | 0.9 | 1.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | pt | |
| -2.0 | -0.1 | 0.7 | 0.8 | 0.8 | 0.8 | 0.9 | 0.9 | -0.1 | -0.1 | -0.1 | -0.1 | 1.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | da | |
| -2.2 | -0.0 | -0.2 | -0.2 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | te | |
| -1.2 | -0.0 | -0.2 | -0.2 | 0.9 | -0.1 | -0.1 | 0.9 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | SV | |
| -1.2 | -0.0 | -0.2 | -0.2 | -0.1 | 0.9 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | no | |
| -1.2 | -0.0 | 1.8 | 0.8 | 1.9 | -0.1 | -0.1 | 0.9 | -0.1 | -0.0 | -0.0 | 1.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | ja | |
| -2.2 | -0.0 | -0.2 | -0.2 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | hi | |
| -0.5 | 1.0 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | sk | |
| 0.5 | -0.0 | 0.9 | 0.9 | 0.9 | -0.1 | -0.1 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 -0.0 | -0.0 | ko | ||
| 0.5 | 1.0 | 1.9 | 1.9 | -0.1 | 0.9 | 0.9 | 1.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 -0.0 | -0.0-0.0 -0.0 -0.0 | he | |||||
| Type | Papers |
| Paper | Survey/PositionDrugan and Babych (2010), Masthoff(2011), Escartin et al. (2(2016), Hovy and Spruit (2O16), Leidner and Plachouras (2017),et al.(2017), Schnoebelen (2017),Larson (2017),De Jong et a(2018),Mayfield et al. (2019),Sun et al. (2019), Blodgett et al.(2021), Bender et al. (2020), Schoch et al. (2020), Savoldi et (2021), Mishra et al. (2021), Czarnowska et al. (2021), DoughSheng et al. (2021), Hovy and Yang (2021), Santy et al. (2021)(2022b), Chandrabose et al. (2021), Vecchi et al. (2021),Talat et :Balkir et al. (2022), Orgad and Belinkov (2022), Akyurek et al(2022), Mohammad (2022a), Orr et al. (2022), Benotti and Blaet al.(2023),Benotti et al. (2023), Ramesh et al. (2023), Rios el |
| Dataset,Benchmark | Yano et al. (2010), Flekova et al. (2016), Braun et al. (2016) KRudinger et al. (2018), Zhao et al. (2018), Kiritchenko and Mo(2019), Gaut et al. (202O),Friedman et al. (2019), Hitti et al. ((2019), Fan et al. (2019),Liu et al. (2020a), Sap et al. (2020),Louet al. (2021), Lim et al. (2020), Shahid et al. (2020), LazaridoUthus (2020), Gala et al. (2020),Dinan et al. (2020b), Meyer e(2020b), Gangula et al. (2019), Zueva et al. (2020), Nangia et al.Blodgett et al. (2021), Spinde et al. (2021), Zhong et al. (2021),Spgaard (2021), Zhao et al. (2021), Aksenov et al. (2021), KamRenduchintala et al. (2021), Barikeri et al. (2021),Milkowski et :et al. (2022), Ziems et al. (2022), Zhou et al. (2022),Akyirek(2022), Bansal et al. (2022),Smith et al. (2022), Qian et al. (2022et al. (2023), Madnani et al. (2017), Kumar and Bhotia (2020)(2021), Zad et al. (2021), Nozza et al. (2022),Mechura (2022) |
| Methods | Gonen and Goldberg (2019), Wevers (2019), Hovy et al. (2020Nissim et al. (2020),Wang etal. (2021d),Wolfe and Caliskan (2et al. (2021), Herold et al. (2022),Bertsch et al. (2022), RampCao et al. (2022a),Alshahrani et al. (2022) Daems and Hackenbuet al. (2015), Terkik et al. (2016), Tatman (2017b),Wang et alTatman (2017a), Zhao et al. (2017), Long et al. (2018), BalyKloppenburg (2019), Farber et al. (2019), Zhong et al. (2019),and Ungar (2019), Gangula et al. (2019), Karve et al. (2019),and Najafian (20i9), Chaloner and Maldonado (2019), Qian et(2019),Kaneko and Bollegala (2019), Huang et al. (2020a),BDinan et al. (2020a), Bordia and Bowman (2O19),Manzini et al, Schwertmann et al. (2023), Pedersen et al. (2023), Bauer et al. (Kumar et al. (2023),Touileb et al. (2023), Parmar et al. (2022),et al. (2023),Alemany et al. (2023), Martinkovä et al. (2023),(2022), Cao et al. (2022b), Sap et al. (2020), Xia et al. (2020(2022), Orgad et al. (2022), Sen et al. (2022), Gupta et al. (202Gira et al. (2022), Guo et al. (2022), Wang et al. (2022), An et a(2022), Borchers et al. (2022),Touileb et al. (2022), Jentzsch andet al. (2022), Tal et al. (2022),Li et al. (2022b),Limisiewicz aret al. (2022),Ahn et al. (2022), Joniak and Aizawa (2022), Chen(2022), Du et al. (2022),Agrawal et al. (2022), Sesari et al. (0Steed et al. (2022), Schick et al. (2021), Zhou et al. (2021), Kaneetal.(2021),Lucy and Bamman (2021), Jin et al.(2021a),)(2021b), Silva et al. (2021), Sun and Peng (2021), Wang et al. (2,Han et al. (2021), Sen et al. (2021),Ramesh et al. (2021), Suband Oh (2021),Davidson et al. (2019),May et al. (2019), SapPiper (2020), Guo et al. (2020), Vu et al. (2020), Spliethöver aMunro and Morrison (2020), Zhao and Chang (2020), Schmahl(2020), Chen et al. (2020b),Mulsa and Spanakis (2020), ShiCotterell(2020),Liang et al. (2020), Wich et al. (2020a), Jia et a(2020),Dev et al. (2021a), Kumar et al. (2020),Field and Tsvetk(2020a), Stafanovics et al. (2020),Li et al. (2020),Ma et al. (AlKuwatly et al. (2020), Bartl et al. (202O),Liu et al. (2020b), I,Sheng et al. (2020), Gaci et al. (2022), Hutchinson et al. (2020)et al. (2021e),Van Der Wal et al. (2022), Hansen and Spgaard (2Azarpanah and Farhadloo (2021), Jiao and Luo (2021), Touileb,Ciora et al. (2021), Murayama et al. (2021), Gaido et al. (202(2022), Dawkins (2021a) , Gillis (2021),Mireshghallah and BWang et al. (2021a),Malik et al. (2022), Subramanian et al. (202Garimella et al. (2021),Ghosh et al. (2021),Mehrabi et al. (20) |
| Both above | Goldfarb-Tarrant et al. (2021), Janghorbani and De Melo (202Troles and Schmid (2021),Liu etal. (2022a),Névéol etal. (202 |
| Analysis | Herzig et al. (2011), Rudinger et al. (2017), Lauscher and GlavaVanderlyn et al. (2021),Falenska and Cetinoglu (2021), Sap et(2021),Bhat et al. (2021), Haroutunian (2022), Qian et al. (201 |
| Year | Papers | Year | Papers |
| 2010-2016 | 11 | 2020 | 68 |
| 2017 | 16 | 2021 | 96 |
| 2018 | 10 | 2022 | 79 |
| 2019 | 46 | 2023 | 22 |
| Totals | 83 | Totals | 265 |
| Year | Papers | Year | Papers |
| 2010-2016 | 11 | 2020 | 68 |
| 2017 | 16 | 2021 | 96 |
| 2018 | 10 | 2022 | 79 |
| 2019 | 46 | 2023 | 22 |
| Totals | 83 | Totals | 265 |
| Type | Papers | % |
| Measurement Debiasing ／Mitigation | 98 52 | 46.66 24.76 |
| Combinations of above | 32 | 15.23 |
| Detection | 19 | 9.04 |
| Generation | 5 | 2.38 |
| Miscellaneous | 4 | 1.9 |
| Total | 210 | 100.0 |
| Type | Papers | % |
| Measurement Debiasing ／Mitigation | 98 52 | 46.66 24.76 |
| Combinations of above | 32 | 15.23 |
| Detection | 19 | 9.04 |
| Generation | 5 | 2.38 |
| Miscellaneous | 4 | 1.9 |
| Total | 210 | 100.0 |
| prrreeade rabrrgsmrs erdnei arrreeeeeingseeenntsuorsseioidnarlgeiroiga[ergiiedJhnrittieaenasrlqesipGrariiagepueeerpeateaaeitueung2 | ||||||||||||||||||
| 19.3 | 8.2 | 8.8 | -8.5 | 5.8 | -8.6 | 3.0 | -6.5 | -7.5 | -6.5 | 5.1 | 1.7 | 3.8 | 3.0 | 1.1 | -2.3 | -1.7 | gender | |
| 19.3 | 12.2 | 12.4 | -1.0 | 6.8 | -3.0 | 4.2 | -1.6 | -2.6 | -2.6 | 8.6 | 3.8 | 4.2 | 4.6 | 0.0 | -0.8 | -0.6 | race | |
| 8.2 | 12.2 | 4.4 | -1.4 | 7.9 | -1.0 | 7.0 | -0.9 | -0.9 | -0.9 | 4.2 | 4.2 | 2.4 | 5.5 | 0.7 | -0.3 | -0.2 | religion | |
| 8.8 | 12.4 | 4.4 | -1.3 | 3.9 | -1.0 | 6.1 | 0.1 | -0.9 | 0.1 | 4.2 | 5.3 | 4.4 | 5.5 | 0.7 | -0.3 | -0.2 | age | |
| -8.5 | -1.0 | -1.4 | -1.3 | 0.1 | 0.1 | 0.2 | -0.7 | -0.7 | -0.7 | -0.7 | -0.6 | -0.5 | -0.4 | -0.3 | -0.2 | 0.8 | sentiment | |
| 5.8 | 6.8 | 7.9 | 3.9 | 0.1 | -0.7 | 5.4 | -0.6 | -0.6 | -0.6 | 3.4 | 4.5 | 0.6 | 4.7 | -0.2 | -0.2 | -0.1 | profession | |
| -8.6 | -3.0 | -1.0 | -1.0 | 0.1 | -0.7 | -0.6 | -0.6 | -0.6 | -0.6 | -0.5 | -0.5 | -0.4 | -0.3 | -0.2 | -0.2 | -0.1 | media | |
| 3.0 | 4.2 | 7.0 | 6.1 | 0.2 | 5.4 | -0.6 | -0.5 | -0.5 | -0.5 | 3.5 | 5.6 | 3.6 | 4.7 | -0.2 | -0.2 | -0.1 | nationality | |
| -6.5 | -1.6 | -0.9 | 0.1 | -0.7 | -0.6 | -0.6 | -0.5 | -0.5 | 0.5 | 0.6 | 0.6 | 0.7 | -0.3 | -0.2 | -0.1 | -0.1 | political | |
| -7.5 | -2.6 | -0.9 | -0.9 | -0.7 | -0.6 | -0.6 | -0.5 | -0.5 | -0.5 | -0.4 | -0.4 | -0.3 | -0.3 | -0.2 | -0.1 | -0.1 | multiplesocial | |
| -6.5 | -2.6 | -0.9 | 0.1 | -0.7 | -0.6 | -0.6 | -0.5 | 0.5 | -0.5 | -0.4 | -0.4 | -0.3 | -0.3 | -0.2 | -0.1 | -0.1 | annotator | |
| 5.1 | 8.6 | 4.2 | 4.2 | -0.7 | 3.4 | -0.5 | 3.5 | 0.6 | -0.4 | -0.4 | 4.6 | 0.7 | 3.8 | -0.2 | -0.1 | -0.1 | sexuality | |
| 1.7 | 3.8 | 4.2 | 5.3 | -0.6 | 4.5 | -0.5 | 5.6 | 0.6 | -0.4 | -0.4 | 4.6 | 1.7 | 4.8 | -0.2 | -0.1 | -0.1 | disability | |
| 3.8 | 4.2 | 2.4 | 4.4 | -0.5 | 0.6 | -0.4 | 3.6 | 0.7 | -0.3 | -0.3 | 0.7 | 1.7 | 0.8 | -0.1 | -0.1 | -0.1 | ethnicity | |
| 3.0 | 4.6 | 5.5 | 5.5 | -0.4 | 4.7 | -0.3 | 4.7 | -0.3 | -0.3 | -0.3 | 3.8 | 4.8 | 0.8 | -0.1 | -0.1 | -0.1 | physicalappearance | |
| 1.1 | 0.0 | 0.7 | 0.7 | -0.3 | -0.2 | -0.2 | -0.2 | -0.2 | -0.2 | -0.2 | -0.2 | -0.2 | -0.1 | -0.1 | -0.1 | -0.0 | dialect | |
| -2.3 | -0.8 | -0.3 | -0.3 | -0.2 | -0.2 | -0.2 | -0.2 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | human | |
| -1.7 | -0.6 | -0.2 | -0.2 | 0.8 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | intersectional | |
| prrreeade rabrrgsmrs erdnei arrreeeeeingseeenntsuorsseioidnarlgeiroiga[ergiiedJhnrittieaenasrlqesipGrariiagepueeerpeateaaeitueung2 | ||||||||||||||||||
| 19.3 | 8.2 | 8.8 | -8.5 | 5.8 | -8.6 | 3.0 | -6.5 | -7.5 | -6.5 | 5.1 | 1.7 | 3.8 | 3.0 | 1.1 | -2.3 | -1.7 | gender | |
| 19.3 | 12.2 | 12.4 | -1.0 | 6.8 | -3.0 | 4.2 | -1.6 | -2.6 | -2.6 | 8.6 | 3.8 | 4.2 | 4.6 | 0.0 | -0.8 | -0.6 | race | |
| 8.2 | 12.2 | 4.4 | -1.4 | 7.9 | -1.0 | 7.0 | -0.9 | -0.9 | -0.9 | 4.2 | 4.2 | 2.4 | 5.5 | 0.7 | -0.3 | -0.2 | religion | |
| 8.8 | 12.4 | 4.4 | -1.3 | 3.9 | -1.0 | 6.1 | 0.1 | -0.9 | 0.1 | 4.2 | 5.3 | 4.4 | 5.5 | 0.7 | -0.3 | -0.2 | age | |
| -8.5 | -1.0 | -1.4 | -1.3 | 0.1 | 0.1 | 0.2 | -0.7 | -0.7 | -0.7 | -0.7 | -0.6 | -0.5 | -0.4 | -0.3 | -0.2 | 0.8 | sentiment | |
| 5.8 | 6.8 | 7.9 | 3.9 | 0.1 | -0.7 | 5.4 | -0.6 | -0.6 | -0.6 | 3.4 | 4.5 | 0.6 | 4.7 | -0.2 | -0.2 | -0.1 | profession | |
| -8.6 | -3.0 | -1.0 | -1.0 | 0.1 | -0.7 | -0.6 | -0.6 | -0.6 | -0.6 | -0.5 | -0.5 | -0.4 | -0.3 | -0.2 | -0.2 | -0.1 | media | |
| 3.0 | 4.2 | 7.0 | 6.1 | 0.2 | 5.4 | -0.6 | -0.5 | -0.5 | -0.5 | 3.5 | 5.6 | 3.6 | 4.7 | -0.2 | -0.2 | -0.1 | nationality | |
| -6.5 | -1.6 | -0.9 | 0.1 | -0.7 | -0.6 | -0.6 | -0.5 | -0.5 | 0.5 | 0.6 | 0.6 | 0.7 | -0.3 | -0.2 | -0.1 | -0.1 | political | |
| -7.5 | -2.6 | -0.9 | -0.9 | -0.7 | -0.6 | -0.6 | -0.5 | -0.5 | -0.5 | -0.4 | -0.4 | -0.3 | -0.3 | -0.2 | -0.1 | -0.1 | multiplesocial | |
| -6.5 | -2.6 | -0.9 | 0.1 | -0.7 | -0.6 | -0.6 | -0.5 | 0.5 | -0.5 | -0.4 | -0.4 | -0.3 | -0.3 | -0.2 | -0.1 | -0.1 | annotator | |
| 5.1 | 8.6 | 4.2 | 4.2 | -0.7 | 3.4 | -0.5 | 3.5 | 0.6 | -0.4 | -0.4 | 4.6 | 0.7 | 3.8 | -0.2 | -0.1 | -0.1 | sexuality | |
| 1.7 | 3.8 | 4.2 | 5.3 | -0.6 | 4.5 | -0.5 | 5.6 | 0.6 | -0.4 | -0.4 | 4.6 | 1.7 | 4.8 | -0.2 | -0.1 | -0.1 | disability | |
| 3.8 | 4.2 | 2.4 | 4.4 | -0.5 | 0.6 | -0.4 | 3.6 | 0.7 | -0.3 | -0.3 | 0.7 | 1.7 | 0.8 | -0.1 | -0.1 | -0.1 | ethnicity | |
| 3.0 | 4.6 | 5.5 | 5.5 | -0.4 | 4.7 | -0.3 | 4.7 | -0.3 | -0.3 | -0.3 | 3.8 | 4.8 | 0.8 | -0.1 | -0.1 | -0.1 | physicalappearance | |
| 1.1 | 0.0 | 0.7 | 0.7 | -0.3 | -0.2 | -0.2 | -0.2 | -0.2 | -0.2 | -0.2 | -0.2 | -0.2 | -0.1 | -0.1 | -0.1 | -0.0 | dialect | |
| -2.3 | -0.8 | -0.3 | -0.3 | -0.2 | -0.2 | -0.2 | -0.2 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | human | |
| -1.7 | -0.6 | -0.2 | -0.2 | 0.8 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | intersectional | |
| en | de | sp | zh1 | fr | it | ru | tr | pl | nl | ar | pt | da | te | SV | no | ja | hi | sk | ko | he | ||
| 0.3 | 2.1 | 2.9 | -4.2 | 1.6 | 0.8 | -1.2 | 1.5 | -0.7 | -1.7 | -0.7 | 0.0 | -2.0 | -2.2 | -1.2 | -1.2 | -1.2 | -2.2 | -0.5 | 0.5 | 0.5 | en | |
| 0.3 | 2.7 | 3.7 | 1.8 | 1.8 | 0.8 | 0.9 | 0.9 | 1.9 | -0.1 | 2.9 | 0.9 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | 1.0 | ||
| 2.1 | 2.7 | 8.7 | 5.0 | 7.0 | 4.2 | 4.5 | 4.6 | 1.7 | 1.7 | 2.7 | 2.7 | 0.7 | -0.2 | -0.2 | -0.2 | 1.8 | -0.2 | -0.1 | 0.9 | 1.9 | de | |
| 2.9 | 3.7 | 8.7 | 3.2 | 6.2 | 5.4 | 3.6 | 3.7 | 1.7 | -0.3 | 3.7 | 3.8 | 0.8 | -0.2 | -0.2 | -0.2 | 0.8 | -0.2 | -0.1 | 0.9 | 1.9 | sp | |
| -4.2 | 1.8 | 5.0 | 3.2 | 1.4 | 0.5 | 1.7 | 1.7 | 0.8 | -0.2 | 2.8 | 1.8 | 0.8 | -0.1 | 0.9 | -0.1 | 1.9 | -0.1 | -0.1 | 0.9 | -0.1 | zh | |
| 1.6 | 1.8 | 7.0 | 6.2 | 1.4 | 6.6 | 1.7 | -0.2 | -0.2 | 0.8 | 1.8 | 0.8 | 0.8 | -0.1 | -0.1 | 0.9 | -0.1 | -0.1 | -0.1 | -0.1 | 0.9 | fr | |
| 0.8 | 0.8 | 4.2 | 5.4 | 0.5 | 6.6 | 1.8 | 0.8 | 0.8 | 0.8 | 0.8 | 1.9 | 0.9 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 -0.1 | -0.1 | 0.9 | it | |
| -1.2 | 0.9 | 4.5 | 3.6 | 1.7 | 1.7 | 1.8 | 1.9 | -0.1 | -0.1 | 1.9 | 0.9 | 0.9 | -0.1 | 0.9 | -0.1 | 0.9 | -0.1 | -0.0 | -0.0 | 1.0 | ru | |
| 1.5 | 0.9 | 4.6 | 3.7 | 1.7 | -0.2 | 0.8 | 1.9 | 1.9 | -0.1 | -0.1 | 0.9 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | 1.0 | -0.0 | tr | |
| -0.7 | 1.9 | 1.7 | 1.7 | 0.8 | -0.2 | 0.8 | -0.1 | 1.9 | -0.1 | -0.1 | 1.9 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | pl | |
| -1.7 | -0.1 | 1.7 | -0.3 | -0.2 | 0.8 | 0.8 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | nl | |
| -0.7 | 2.9 | 2.7 | 3.7 | 2.8 | 1.8 | 0.8 | 1.9 | -0.1 | -0.1 | -0.1 | 0.9 | -0.1 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | 1.0 | ar | |
| 0.0 | 0.9 | 2.7 | 3.8 | 1.8 | 0.8 | 1.9 | 0.9 | 0.9 | 1.9 | -0.1 | 0.9 | 1.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | pt | |
| -2.0 | -0.1 | 0.7 | 0.8 | 0.8 | 0.8 | 0.9 | 0.9 | -0.1 | -0.1 | -0.1 | -0.1 | 1.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | da | |
| -2.2 | -0.0 | -0.2 | -0.2 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | te | |
| -1.2 | -0.0 | -0.2 | -0.2 | 0.9 | -0.1 | -0.1 | 0.9 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | SV | |
| -1.2 | -0.0 | -0.2 | -0.2 | -0.1 | 0.9 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | no | |
| -1.2 | -0.0 | 1.8 | 0.8 | 1.9 | -0.1 | -0.1 | 0.9 | -0.1 | -0.0 | -0.0 | 1.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | ja | |
| -2.2 | -0.0 | -0.2 | -0.2 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | hi | |
| -0.5 | 1.0 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | sk | |
| 0.5 | -0.0 | 0.9 | 0.9 | 0.9 | -0.1 | -0.1 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 -0.0 | -0.0 | ko | ||
| 0.5 | 1.0 | 1.9 | 1.9 | -0.1 | 0.9 | 0.9 | 1.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 -0.0 | -0.0-0.0 -0.0 -0.0 | he | |||||
| en | de | sp | zh1 | fr | it | ru | tr | pl | nl | ar | pt | da | te | SV | no | ja | hi | sk | ko | he | ||
| 0.3 | 2.1 | 2.9 | -4.2 | 1.6 | 0.8 | -1.2 | 1.5 | -0.7 | -1.7 | -0.7 | 0.0 | -2.0 | -2.2 | -1.2 | -1.2 | -1.2 | -2.2 | -0.5 | 0.5 | 0.5 | en | |
| 0.3 | 2.7 | 3.7 | 1.8 | 1.8 | 0.8 | 0.9 | 0.9 | 1.9 | -0.1 | 2.9 | 0.9 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | 1.0 | ||
| 2.1 | 2.7 | 8.7 | 5.0 | 7.0 | 4.2 | 4.5 | 4.6 | 1.7 | 1.7 | 2.7 | 2.7 | 0.7 | -0.2 | -0.2 | -0.2 | 1.8 | -0.2 | -0.1 | 0.9 | 1.9 | de | |
| 2.9 | 3.7 | 8.7 | 3.2 | 6.2 | 5.4 | 3.6 | 3.7 | 1.7 | -0.3 | 3.7 | 3.8 | 0.8 | -0.2 | -0.2 | -0.2 | 0.8 | -0.2 | -0.1 | 0.9 | 1.9 | sp | |
| -4.2 | 1.8 | 5.0 | 3.2 | 1.4 | 0.5 | 1.7 | 1.7 | 0.8 | -0.2 | 2.8 | 1.8 | 0.8 | -0.1 | 0.9 | -0.1 | 1.9 | -0.1 | -0.1 | 0.9 | -0.1 | zh | |
| 1.6 | 1.8 | 7.0 | 6.2 | 1.4 | 6.6 | 1.7 | -0.2 | -0.2 | 0.8 | 1.8 | 0.8 | 0.8 | -0.1 | -0.1 | 0.9 | -0.1 | -0.1 | -0.1 | -0.1 | 0.9 | fr | |
| 0.8 | 0.8 | 4.2 | 5.4 | 0.5 | 6.6 | 1.8 | 0.8 | 0.8 | 0.8 | 0.8 | 1.9 | 0.9 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 -0.1 | -0.1 | 0.9 | it | |
| -1.2 | 0.9 | 4.5 | 3.6 | 1.7 | 1.7 | 1.8 | 1.9 | -0.1 | -0.1 | 1.9 | 0.9 | 0.9 | -0.1 | 0.9 | -0.1 | 0.9 | -0.1 | -0.0 | -0.0 | 1.0 | ru | |
| 1.5 | 0.9 | 4.6 | 3.7 | 1.7 | -0.2 | 0.8 | 1.9 | 1.9 | -0.1 | -0.1 | 0.9 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | 1.0 | -0.0 | tr | |
| -0.7 | 1.9 | 1.7 | 1.7 | 0.8 | -0.2 | 0.8 | -0.1 | 1.9 | -0.1 | -0.1 | 1.9 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | pl | |
| -1.7 | -0.1 | 1.7 | -0.3 | -0.2 | 0.8 | 0.8 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | nl | |
| -0.7 | 2.9 | 2.7 | 3.7 | 2.8 | 1.8 | 0.8 | 1.9 | -0.1 | -0.1 | -0.1 | 0.9 | -0.1 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | 1.0 | ar | |
| 0.0 | 0.9 | 2.7 | 3.8 | 1.8 | 0.8 | 1.9 | 0.9 | 0.9 | 1.9 | -0.1 | 0.9 | 1.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | pt | |
| -2.0 | -0.1 | 0.7 | 0.8 | 0.8 | 0.8 | 0.9 | 0.9 | -0.1 | -0.1 | -0.1 | -0.1 | 1.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | da | |
| -2.2 | -0.0 | -0.2 | -0.2 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | te | |
| -1.2 | -0.0 | -0.2 | -0.2 | 0.9 | -0.1 | -0.1 | 0.9 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | SV | |
| -1.2 | -0.0 | -0.2 | -0.2 | -0.1 | 0.9 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | no | |
| -1.2 | -0.0 | 1.8 | 0.8 | 1.9 | -0.1 | -0.1 | 0.9 | -0.1 | -0.0 | -0.0 | 1.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | ja | |
| -2.2 | -0.0 | -0.2 | -0.2 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | hi | |
| -0.5 | 1.0 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | sk | |
| 0.5 | -0.0 | 0.9 | 0.9 | 0.9 | -0.1 | -0.1 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 -0.0 | -0.0 | ko | ||
| 0.5 | 1.0 | 1.9 | 1.9 | -0.1 | 0.9 | 0.9 | 1.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 -0.0 | -0.0-0.0 -0.0 -0.0 | he | |||||
| Type | Papers |
| Paper | Survey/PositionDrugan and Babych (2010), Masthoff(2011), Escartin et al. (2(2016), Hovy and Spruit (2O16), Leidner and Plachouras (2017),et al.(2017), Schnoebelen (2017),Larson (2017),De Jong et a(2018),Mayfield et al. (2019),Sun et al. (2019), Blodgett et al.(2021), Bender et al. (2020), Schoch et al. (2020), Savoldi et (2021), Mishra et al. (2021), Czarnowska et al. (2021), DoughSheng et al. (2021), Hovy and Yang (2021), Santy et al. (2021)(2022b), Chandrabose et al. (2021), Vecchi et al. (2021),Talat et :Balkir et al. (2022), Orgad and Belinkov (2022), Akyurek et al(2022), Mohammad (2022a), Orr et al. (2022), Benotti and Blaet al.(2023),Benotti et al. (2023), Ramesh et al. (2023), Rios el |
| Dataset,Benchmark | Yano et al. (2010), Flekova et al. (2016), Braun et al. (2016) KRudinger et al. (2018), Zhao et al. (2018), Kiritchenko and Mo(2019), Gaut et al. (202O),Friedman et al. (2019), Hitti et al. ((2019), Fan et al. (2019),Liu et al. (2020a), Sap et al. (2020),Louet al. (2021), Lim et al. (2020), Shahid et al. (2020), LazaridoUthus (2020), Gala et al. (2020),Dinan et al. (2020b), Meyer e(2020b), Gangula et al. (2019), Zueva et al. (2020), Nangia et al.Blodgett et al. (2021), Spinde et al. (2021), Zhong et al. (2021),Spgaard (2021), Zhao et al. (2021), Aksenov et al. (2021), KamRenduchintala et al. (2021), Barikeri et al. (2021),Milkowski et :et al. (2022), Ziems et al. (2022), Zhou et al. (2022),Akyirek(2022), Bansal et al. (2022),Smith et al. (2022), Qian et al. (2022et al. (2023), Madnani et al. (2017), Kumar and Bhotia (2020)(2021), Zad et al. (2021), Nozza et al. (2022),Mechura (2022) |
| Methods | Gonen and Goldberg (2019), Wevers (2019), Hovy et al. (2020Nissim et al. (2020),Wang etal. (2021d),Wolfe and Caliskan (2et al. (2021), Herold et al. (2022),Bertsch et al. (2022), RampCao et al. (2022a),Alshahrani et al. (2022) Daems and Hackenbuet al. (2015), Terkik et al. (2016), Tatman (2017b),Wang et alTatman (2017a), Zhao et al. (2017), Long et al. (2018), BalyKloppenburg (2019), Farber et al. (2019), Zhong et al. (2019),and Ungar (2019), Gangula et al. (2019), Karve et al. (2019),and Najafian (20i9), Chaloner and Maldonado (2019), Qian et(2019),Kaneko and Bollegala (2019), Huang et al. (2020a),BDinan et al. (2020a), Bordia and Bowman (2O19),Manzini et al, Schwertmann et al. (2023), Pedersen et al. (2023), Bauer et al. (Kumar et al. (2023),Touileb et al. (2023), Parmar et al. (2022),et al. (2023),Alemany et al. (2023), Martinkovä et al. (2023),(2022), Cao et al. (2022b), Sap et al. (2020), Xia et al. (2020(2022), Orgad et al. (2022), Sen et al. (2022), Gupta et al. (202Gira et al. (2022), Guo et al. (2022), Wang et al. (2022), An et a(2022), Borchers et al. (2022),Touileb et al. (2022), Jentzsch andet al. (2022), Tal et al. (2022),Li et al. (2022b),Limisiewicz aret al. (2022),Ahn et al. (2022), Joniak and Aizawa (2022), Chen(2022), Du et al. (2022),Agrawal et al. (2022), Sesari et al. (0Steed et al. (2022), Schick et al. (2021), Zhou et al. (2021), Kaneetal.(2021),Lucy and Bamman (2021), Jin et al.(2021a),)(2021b), Silva et al. (2021), Sun and Peng (2021), Wang et al. (2,Han et al. (2021), Sen et al. (2021),Ramesh et al. (2021), Suband Oh (2021),Davidson et al. (2019),May et al. (2019), SapPiper (2020), Guo et al. (2020), Vu et al. (2020), Spliethöver aMunro and Morrison (2020), Zhao and Chang (2020), Schmahl(2020), Chen et al. (2020b),Mulsa and Spanakis (2020), ShiCotterell(2020),Liang et al. (2020), Wich et al. (2020a), Jia et a(2020),Dev et al. (2021a), Kumar et al. (2020),Field and Tsvetk(2020a), Stafanovics et al. (2020),Li et al. (2020),Ma et al. (AlKuwatly et al. (2020), Bartl et al. (202O),Liu et al. (2020b), I,Sheng et al. (2020), Gaci et al. (2022), Hutchinson et al. (2020)et al. (2021e),Van Der Wal et al. (2022), Hansen and Spgaard (2Azarpanah and Farhadloo (2021), Jiao and Luo (2021), Touileb,Ciora et al. (2021), Murayama et al. (2021), Gaido et al. (202(2022), Dawkins (2021a) , Gillis (2021),Mireshghallah and BWang et al. (2021a),Malik et al. (2022), Subramanian et al. (202Garimella et al. (2021),Ghosh et al. (2021),Mehrabi et al. (20) |
| Both above | Goldfarb-Tarrant et al. (2021), Janghorbani and De Melo (202Troles and Schmid (2021),Liu etal. (2022a),Névéol etal. (202 |
| Analysis | Herzig et al. (2011), Rudinger et al. (2017), Lauscher and GlavaVanderlyn et al. (2021),Falenska and Cetinoglu (2021), Sap et(2021),Bhat et al. (2021), Haroutunian (2022), Qian et al. (201 |
| Type | Papers |
| Paper | Survey/PositionDrugan and Babych (2010), Masthoff(2011), Escartin et al. (2(2016), Hovy and Spruit (2O16), Leidner and Plachouras (2017),et al.(2017), Schnoebelen (2017),Larson (2017),De Jong et a(2018),Mayfield et al. (2019),Sun et al. (2019), Blodgett et al.(2021), Bender et al. (2020), Schoch et al. (2020), Savoldi et (2021), Mishra et al. (2021), Czarnowska et al. (2021), DoughSheng et al. (2021), Hovy and Yang (2021), Santy et al. (2021)(2022b), Chandrabose et al. (2021), Vecchi et al. (2021),Talat et :Balkir et al. (2022), Orgad and Belinkov (2022), Akyurek et al(2022), Mohammad (2022a), Orr et al. (2022), Benotti and Blaet al.(2023),Benotti et al. (2023), Ramesh et al. (2023), Rios el |
| Dataset,Benchmark | Yano et al. (2010), Flekova et al. (2016), Braun et al. (2016) KRudinger et al. (2018), Zhao et al. (2018), Kiritchenko and Mo(2019), Gaut et al. (202O),Friedman et al. (2019), Hitti et al. ((2019), Fan et al. (2019),Liu et al. (2020a), Sap et al. (2020),Louet al. (2021), Lim et al. (2020), Shahid et al. (2020), LazaridoUthus (2020), Gala et al. (2020),Dinan et al. (2020b), Meyer e(2020b), Gangula et al. (2019), Zueva et al. (2020), Nangia et al.Blodgett et al. (2021), Spinde et al. (2021), Zhong et al. (2021),Spgaard (2021), Zhao et al. (2021), Aksenov et al. (2021), KamRenduchintala et al. (2021), Barikeri et al. (2021),Milkowski et :et al. (2022), Ziems et al. (2022), Zhou et al. (2022),Akyirek(2022), Bansal et al. (2022),Smith et al. (2022), Qian et al. (2022et al. (2023), Madnani et al. (2017), Kumar and Bhotia (2020)(2021), Zad et al. (2021), Nozza et al. (2022),Mechura (2022) |
| Methods | Gonen and Goldberg (2019), Wevers (2019), Hovy et al. (2020Nissim et al. (2020),Wang etal. (2021d),Wolfe and Caliskan (2et al. (2021), Herold et al. (2022),Bertsch et al. (2022), RampCao et al. (2022a),Alshahrani et al. (2022) Daems and Hackenbuet al. (2015), Terkik et al. (2016), Tatman (2017b),Wang et alTatman (2017a), Zhao et al. (2017), Long et al. (2018), BalyKloppenburg (2019), Farber et al. (2019), Zhong et al. (2019),and Ungar (2019), Gangula et al. (2019), Karve et al. (2019),and Najafian (20i9), Chaloner and Maldonado (2019), Qian et(2019),Kaneko and Bollegala (2019), Huang et al. (2020a),BDinan et al. (2020a), Bordia and Bowman (2O19),Manzini et al, Schwertmann et al. (2023), Pedersen et al. (2023), Bauer et al. (Kumar et al. (2023),Touileb et al. (2023), Parmar et al. (2022),et al. (2023),Alemany et al. (2023), Martinkovä et al. (2023),(2022), Cao et al. (2022b), Sap et al. (2020), Xia et al. (2020(2022), Orgad et al. (2022), Sen et al. (2022), Gupta et al. (202Gira et al. (2022), Guo et al. (2022), Wang et al. (2022), An et a(2022), Borchers et al. (2022),Touileb et al. (2022), Jentzsch andet al. (2022), Tal et al. (2022),Li et al. (2022b),Limisiewicz aret al. (2022),Ahn et al. (2022), Joniak and Aizawa (2022), Chen(2022), Du et al. (2022),Agrawal et al. (2022), Sesari et al. (0Steed et al. (2022), Schick et al. (2021), Zhou et al. (2021), Kaneetal.(2021),Lucy and Bamman (2021), Jin et al.(2021a),)(2021b), Silva et al. (2021), Sun and Peng (2021), Wang et al. (2,Han et al. (2021), Sen et al. (2021),Ramesh et al. (2021), Suband Oh (2021),Davidson et al. (2019),May et al. (2019), SapPiper (2020), Guo et al. (2020), Vu et al. (2020), Spliethöver aMunro and Morrison (2020), Zhao and Chang (2020), Schmahl(2020), Chen et al. (2020b),Mulsa and Spanakis (2020), ShiCotterell(2020),Liang et al. (2020), Wich et al. (2020a), Jia et a(2020),Dev et al. (2021a), Kumar et al. (2020),Field and Tsvetk(2020a), Stafanovics et al. (2020),Li et al. (2020),Ma et al. (AlKuwatly et al. (2020), Bartl et al. (202O),Liu et al. (2020b), I,Sheng et al. (2020), Gaci et al. (2022), Hutchinson et al. (2020)et al. (2021e),Van Der Wal et al. (2022), Hansen and Spgaard (2Azarpanah and Farhadloo (2021), Jiao and Luo (2021), Touileb,Ciora et al. (2021), Murayama et al. (2021), Gaido et al. (202(2022), Dawkins (2021a) , Gillis (2021),Mireshghallah and BWang et al. (2021a),Malik et al. (2022), Subramanian et al. (202Garimella et al. (2021),Ghosh et al. (2021),Mehrabi et al. (20) |
| Both above | Goldfarb-Tarrant et al. (2021), Janghorbani and De Melo (202Troles and Schmid (2021),Liu etal. (2022a),Névéol etal. (202 |
| Analysis | Herzig et al. (2011), Rudinger et al. (2017), Lauscher and GlavaVanderlyn et al. (2021),Falenska and Cetinoglu (2021), Sap et(2021),Bhat et al. (2021), Haroutunian (2022), Qian et al. (201 |
| Year | Papers | Year | Papers |
| 2010-2016 | 11 | 2020 | 68 |
| 2017 | 16 | 2021 | 96 |
| 2018 | 10 | 2022 | 79 |
| 2019 | 46 | 2023 | 22 |
| Totals | 83 | Totals | 265 |
| Type | Papers | % |
| Measurement Debiasing ／Mitigation | 98 52 | 46.66 24.76 |
| Combinations of above | 32 | 15.23 |
| Detection | 19 | 9.04 |
| Generation | 5 | 2.38 |
| Miscellaneous | 4 | 1.9 |
| Total | 210 | 100.0 |
| prrreeade rabrrgsmrs erdnei arrreeeeeingseeenntsuorsseioidnarlgeiroiga[ergiiedJhnrittieaenasrlqesipGrariiagepueeerpeateaaeitueung2 | ||||||||||||||||||
| 19.3 | 8.2 | 8.8 | -8.5 | 5.8 | -8.6 | 3.0 | -6.5 | -7.5 | -6.5 | 5.1 | 1.7 | 3.8 | 3.0 | 1.1 | -2.3 | -1.7 | gender | |
| 19.3 | 12.2 | 12.4 | -1.0 | 6.8 | -3.0 | 4.2 | -1.6 | -2.6 | -2.6 | 8.6 | 3.8 | 4.2 | 4.6 | 0.0 | -0.8 | -0.6 | race | |
| 8.2 | 12.2 | 4.4 | -1.4 | 7.9 | -1.0 | 7.0 | -0.9 | -0.9 | -0.9 | 4.2 | 4.2 | 2.4 | 5.5 | 0.7 | -0.3 | -0.2 | religion | |
| 8.8 | 12.4 | 4.4 | -1.3 | 3.9 | -1.0 | 6.1 | 0.1 | -0.9 | 0.1 | 4.2 | 5.3 | 4.4 | 5.5 | 0.7 | -0.3 | -0.2 | age | |
| -8.5 | -1.0 | -1.4 | -1.3 | 0.1 | 0.1 | 0.2 | -0.7 | -0.7 | -0.7 | -0.7 | -0.6 | -0.5 | -0.4 | -0.3 | -0.2 | 0.8 | sentiment | |
| 5.8 | 6.8 | 7.9 | 3.9 | 0.1 | -0.7 | 5.4 | -0.6 | -0.6 | -0.6 | 3.4 | 4.5 | 0.6 | 4.7 | -0.2 | -0.2 | -0.1 | profession | |
| -8.6 | -3.0 | -1.0 | -1.0 | 0.1 | -0.7 | -0.6 | -0.6 | -0.6 | -0.6 | -0.5 | -0.5 | -0.4 | -0.3 | -0.2 | -0.2 | -0.1 | media | |
| 3.0 | 4.2 | 7.0 | 6.1 | 0.2 | 5.4 | -0.6 | -0.5 | -0.5 | -0.5 | 3.5 | 5.6 | 3.6 | 4.7 | -0.2 | -0.2 | -0.1 | nationality | |
| -6.5 | -1.6 | -0.9 | 0.1 | -0.7 | -0.6 | -0.6 | -0.5 | -0.5 | 0.5 | 0.6 | 0.6 | 0.7 | -0.3 | -0.2 | -0.1 | -0.1 | political | |
| -7.5 | -2.6 | -0.9 | -0.9 | -0.7 | -0.6 | -0.6 | -0.5 | -0.5 | -0.5 | -0.4 | -0.4 | -0.3 | -0.3 | -0.2 | -0.1 | -0.1 | multiplesocial | |
| -6.5 | -2.6 | -0.9 | 0.1 | -0.7 | -0.6 | -0.6 | -0.5 | 0.5 | -0.5 | -0.4 | -0.4 | -0.3 | -0.3 | -0.2 | -0.1 | -0.1 | annotator | |
| 5.1 | 8.6 | 4.2 | 4.2 | -0.7 | 3.4 | -0.5 | 3.5 | 0.6 | -0.4 | -0.4 | 4.6 | 0.7 | 3.8 | -0.2 | -0.1 | -0.1 | sexuality | |
| 1.7 | 3.8 | 4.2 | 5.3 | -0.6 | 4.5 | -0.5 | 5.6 | 0.6 | -0.4 | -0.4 | 4.6 | 1.7 | 4.8 | -0.2 | -0.1 | -0.1 | disability | |
| 3.8 | 4.2 | 2.4 | 4.4 | -0.5 | 0.6 | -0.4 | 3.6 | 0.7 | -0.3 | -0.3 | 0.7 | 1.7 | 0.8 | -0.1 | -0.1 | -0.1 | ethnicity | |
| 3.0 | 4.6 | 5.5 | 5.5 | -0.4 | 4.7 | -0.3 | 4.7 | -0.3 | -0.3 | -0.3 | 3.8 | 4.8 | 0.8 | -0.1 | -0.1 | -0.1 | physicalappearance | |
| 1.1 | 0.0 | 0.7 | 0.7 | -0.3 | -0.2 | -0.2 | -0.2 | -0.2 | -0.2 | -0.2 | -0.2 | -0.2 | -0.1 | -0.1 | -0.1 | -0.0 | dialect | |
| -2.3 | -0.8 | -0.3 | -0.3 | -0.2 | -0.2 | -0.2 | -0.2 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | human | |
| -1.7 | -0.6 | -0.2 | -0.2 | 0.8 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | intersectional | |
| en | de | sp | zh1 | fr | it | ru | tr | pl | nl | ar | pt | da | te | SV | no | ja | hi | sk | ko | he | ||
| 0.3 | 2.1 | 2.9 | -4.2 | 1.6 | 0.8 | -1.2 | 1.5 | -0.7 | -1.7 | -0.7 | 0.0 | -2.0 | -2.2 | -1.2 | -1.2 | -1.2 | -2.2 | -0.5 | 0.5 | 0.5 | en | |
| 0.3 | 2.7 | 3.7 | 1.8 | 1.8 | 0.8 | 0.9 | 0.9 | 1.9 | -0.1 | 2.9 | 0.9 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | 1.0 | ||
| 2.1 | 2.7 | 8.7 | 5.0 | 7.0 | 4.2 | 4.5 | 4.6 | 1.7 | 1.7 | 2.7 | 2.7 | 0.7 | -0.2 | -0.2 | -0.2 | 1.8 | -0.2 | -0.1 | 0.9 | 1.9 | de | |
| 2.9 | 3.7 | 8.7 | 3.2 | 6.2 | 5.4 | 3.6 | 3.7 | 1.7 | -0.3 | 3.7 | 3.8 | 0.8 | -0.2 | -0.2 | -0.2 | 0.8 | -0.2 | -0.1 | 0.9 | 1.9 | sp | |
| -4.2 | 1.8 | 5.0 | 3.2 | 1.4 | 0.5 | 1.7 | 1.7 | 0.8 | -0.2 | 2.8 | 1.8 | 0.8 | -0.1 | 0.9 | -0.1 | 1.9 | -0.1 | -0.1 | 0.9 | -0.1 | zh | |
| 1.6 | 1.8 | 7.0 | 6.2 | 1.4 | 6.6 | 1.7 | -0.2 | -0.2 | 0.8 | 1.8 | 0.8 | 0.8 | -0.1 | -0.1 | 0.9 | -0.1 | -0.1 | -0.1 | -0.1 | 0.9 | fr | |
| 0.8 | 0.8 | 4.2 | 5.4 | 0.5 | 6.6 | 1.8 | 0.8 | 0.8 | 0.8 | 0.8 | 1.9 | 0.9 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 -0.1 | -0.1 | 0.9 | it | |
| -1.2 | 0.9 | 4.5 | 3.6 | 1.7 | 1.7 | 1.8 | 1.9 | -0.1 | -0.1 | 1.9 | 0.9 | 0.9 | -0.1 | 0.9 | -0.1 | 0.9 | -0.1 | -0.0 | -0.0 | 1.0 | ru | |
| 1.5 | 0.9 | 4.6 | 3.7 | 1.7 | -0.2 | 0.8 | 1.9 | 1.9 | -0.1 | -0.1 | 0.9 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | 1.0 | -0.0 | tr | |
| -0.7 | 1.9 | 1.7 | 1.7 | 0.8 | -0.2 | 0.8 | -0.1 | 1.9 | -0.1 | -0.1 | 1.9 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | pl | |
| -1.7 | -0.1 | 1.7 | -0.3 | -0.2 | 0.8 | 0.8 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | nl | |
| -0.7 | 2.9 | 2.7 | 3.7 | 2.8 | 1.8 | 0.8 | 1.9 | -0.1 | -0.1 | -0.1 | 0.9 | -0.1 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | 1.0 | ar | |
| 0.0 | 0.9 | 2.7 | 3.8 | 1.8 | 0.8 | 1.9 | 0.9 | 0.9 | 1.9 | -0.1 | 0.9 | 1.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | pt | |
| -2.0 | -0.1 | 0.7 | 0.8 | 0.8 | 0.8 | 0.9 | 0.9 | -0.1 | -0.1 | -0.1 | -0.1 | 1.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | da | |
| -2.2 | -0.0 | -0.2 | -0.2 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | te | |
| -1.2 | -0.0 | -0.2 | -0.2 | 0.9 | -0.1 | -0.1 | 0.9 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | SV | |
| -1.2 | -0.0 | -0.2 | -0.2 | -0.1 | 0.9 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | no | |
| -1.2 | -0.0 | 1.8 | 0.8 | 1.9 | -0.1 | -0.1 | 0.9 | -0.1 | -0.0 | -0.0 | 1.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | ja | |
| -2.2 | -0.0 | -0.2 | -0.2 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | hi | |
| -0.5 | 1.0 | -0.1 | -0.1 | -0.1 | -0.1 | -0.1 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | sk | |
| 0.5 | -0.0 | 0.9 | 0.9 | 0.9 | -0.1 | -0.1 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 -0.0 | -0.0 | ko | ||
| 0.5 | 1.0 | 1.9 | 1.9 | -0.1 | 0.9 | 0.9 | 1.0 | -0.0 | -0.0 | -0.0 | 1.0 | -0.0 | -0.0 | -0.0 | -0.0 -0.0 | -0.0-0.0 -0.0 -0.0 | he | |||||
| Type | Papers |
| Paper | Survey/PositionDrugan and Babych (2010), Masthoff(2011), Escartin et al. (2(2016), Hovy and Spruit (2O16), Leidner and Plachouras (2017),et al.(2017), Schnoebelen (2017),Larson (2017),De Jong et a(2018),Mayfield et al. (2019),Sun et al. (2019), Blodgett et al.(2021), Bender et al. (2020), Schoch et al. (2020), Savoldi et (2021), Mishra et al. (2021), Czarnowska et al. (2021), DoughSheng et al. (2021), Hovy and Yang (2021), Santy et al. (2021)(2022b), Chandrabose et al. (2021), Vecchi et al. (2021),Talat et :Balkir et al. (2022), Orgad and Belinkov (2022), Akyurek et al(2022), Mohammad (2022a), Orr et al. (2022), Benotti and Blaet al.(2023),Benotti et al. (2023), Ramesh et al. (2023), Rios el |
| Dataset,Benchmark | Yano et al. (2010), Flekova et al. (2016), Braun et al. (2016) KRudinger et al. (2018), Zhao et al. (2018), Kiritchenko and Mo(2019), Gaut et al. (202O),Friedman et al. (2019), Hitti et al. ((2019), Fan et al. (2019),Liu et al. (2020a), Sap et al. (2020),Louet al. (2021), Lim et al. (2020), Shahid et al. (2020), LazaridoUthus (2020), Gala et al. (2020),Dinan et al. (2020b), Meyer e(2020b), Gangula et al. (2019), Zueva et al. (2020), Nangia et al.Blodgett et al. (2021), Spinde et al. (2021), Zhong et al. (2021),Spgaard (2021), Zhao et al. (2021), Aksenov et al. (2021), KamRenduchintala et al. (2021), Barikeri et al. (2021),Milkowski et :et al. (2022), Ziems et al. (2022), Zhou et al. (2022),Akyirek(2022), Bansal et al. (2022),Smith et al. (2022), Qian et al. (2022et al. (2023), Madnani et al. (2017), Kumar and Bhotia (2020)(2021), Zad et al. (2021), Nozza et al. (2022),Mechura (2022) |
| Methods | Gonen and Goldberg (2019), Wevers (2019), Hovy et al. (2020Nissim et al. (2020),Wang etal. (2021d),Wolfe and Caliskan (2et al. (2021), Herold et al. (2022),Bertsch et al. (2022), RampCao et al. (2022a),Alshahrani et al. (2022) Daems and Hackenbuet al. (2015), Terkik et al. (2016), Tatman (2017b),Wang et alTatman (2017a), Zhao et al. (2017), Long et al. (2018), BalyKloppenburg (2019), Farber et al. (2019), Zhong et al. (2019),and Ungar (2019), Gangula et al. (2019), Karve et al. (2019),and Najafian (20i9), Chaloner and Maldonado (2019), Qian et(2019),Kaneko and Bollegala (2019), Huang et al. (2020a),BDinan et al. (2020a), Bordia and Bowman (2O19),Manzini et al, Schwertmann et al. (2023), Pedersen et al. (2023), Bauer et al. (Kumar et al. (2023),Touileb et al. (2023), Parmar et al. (2022),et al. (2023),Alemany et al. (2023), Martinkovä et al. (2023),(2022), Cao et al. (2022b), Sap et al. (2020), Xia et al. (2020(2022), Orgad et al. (2022), Sen et al. (2022), Gupta et al. (202Gira et al. (2022), Guo et al. (2022), Wang et al. (2022), An et a(2022), Borchers et al. (2022),Touileb et al. (2022), Jentzsch andet al. (2022), Tal et al. (2022),Li et al. (2022b),Limisiewicz aret al. (2022),Ahn et al. (2022), Joniak and Aizawa (2022), Chen(2022), Du et al. (2022),Agrawal et al. (2022), Sesari et al. (0Steed et al. (2022), Schick et al. (2021), Zhou et al. (2021), Kaneetal.(2021),Lucy and Bamman (2021), Jin et al.(2021a),)(2021b), Silva et al. (2021), Sun and Peng (2021), Wang et al. (2,Han et al. (2021), Sen et al. (2021),Ramesh et al. (2021), Suband Oh (2021),Davidson et al. (2019),May et al. (2019), SapPiper (2020), Guo et al. (2020), Vu et al. (2020), Spliethöver aMunro and Morrison (2020), Zhao and Chang (2020), Schmahl(2020), Chen et al. (2020b),Mulsa and Spanakis (2020), ShiCotterell(2020),Liang et al. (2020), Wich et al. (2020a), Jia et a(2020),Dev et al. (2021a), Kumar et al. (2020),Field and Tsvetk(2020a), Stafanovics et al. (2020),Li et al. (2020),Ma et al. (AlKuwatly et al. (2020), Bartl et al. (202O),Liu et al. (2020b), I,Sheng et al. (2020), Gaci et al. (2022), Hutchinson et al. (2020)et al. (2021e),Van Der Wal et al. (2022), Hansen and Spgaard (2Azarpanah and Farhadloo (2021), Jiao and Luo (2021), Touileb,Ciora et al. (2021), Murayama et al. (2021), Gaido et al. (202(2022), Dawkins (2021a) , Gillis (2021),Mireshghallah and BWang et al. (2021a),Malik et al. (2022), Subramanian et al. (202Garimella et al. (2021),Ghosh et al. (2021),Mehrabi et al. (20) |
| Both above | Goldfarb-Tarrant et al. (2021), Janghorbani and De Melo (202Troles and Schmid (2021),Liu etal. (2022a),Névéol etal. (202 |
| Analysis | Herzig et al. (2011), Rudinger et al. (2017), Lauscher and GlavaVanderlyn et al. (2021),Falenska and Cetinoglu (2021), Sap et(2021),Bhat et al. (2021), Haroutunian (2022), Qian et al. (201 |
| CelebA-HQ in-domain | CelebA-HQ unseen-domain | LSUN-church | ||||||||
| quality | attribute | overall | quality | attribute | overall | quality | attribute | diversity | overall | |
| Asyrp (ours) | 98.36% | 88.13% | 94.92% | 71.56% | 59.84% | 63.13% | 73.19% | 71.81% | 87.50% | 76.81% |
| DiffusionCLIP | 1.64% | 11.88% | 5.08% | 28.44% | 40.16% | 36.88% | 26.81% | 28.19% | 12.50% | 23.19% |
| Dataset | src_txt | trg_txt | CLIP similarity | tedit | tboost |
| CelebA-HQ LSUN-church | Person Face | Young person | 0.905 | 515 | 167 |
| Smiling face | 0.899 | 513 | 167 | ||
| Face | Sad face | 0.894 | 513 | 167 | |
| Face | Angry face | 0.892 | 512 | 167 | |
| Face | Tanned face | 0.886 | 512 | 167 | |
| Face | Disgusted face | 0.880 | 511 | 167 | |
| Person Human | Person with makeup | 0.875 | 509 | 167 | |
| Person | Zombie | 0.868 | 506 | 167 | |
| Person | Person with bald head | 0.861 | 505 | 167 | |
| Person with curly hair | 0.835 | 499 | 167 | ||
| Human Person | Neanderthal | 0.802 | 490 | 167 | |
| Person | Mark Zuckerberg | 0.797 | 489 | 167 | |
| Human | Nicolas Cage | 0.710 | 461 | 167 | |
| Photo | Painting in the style of Pixar | 0.667 | 446 | 167 | |
| Photo | Painting in Modigliani style Self-portrait by Frida Kahlo | 0.565 | 403 | 167 | |
| Church | Gothic Church | 0.443 | 321 | 167 | |
| Church | Temple | 0.912 0.898 | 371 | 293 | |
| Church | Department store | 0.841 | 367 | 293 | |
| Church | Wooden House | 349 | 293 | ||
| Church | 0.793 | 333 | 293 | ||
| Church | Ancient traditional Asian tower | 0.784 | 330 | 293 | |
| Church | Red brick wall Church | 0.774 | 326 | 293 | |
| Bedroom | Factory Princess Bedroom | 0.702 | 301 | 293 | |
| LSUN-bedroom | Bedroom Hotel Bedroom | 0.912 | 370 | 221 | |
| AFHQ | Dog | 0.917 | 371 | 221 | |
| Dog | Happy Dog | 0.883 | 430 | 167 | |
| Sleepy Dog | 0.866 | 422 | 167 | ||
| Dog | Angry Dog | 0.860 | 419 | 167 | |
| Dog Dog | Wolf | 0.850 | 412 | 167 | |
| METFACES | Painting of a person | Yorkshire Terrier | 0.690 | 302 | 167 |
| Painting of a person | Painting of a Sad person Painting of a Smiling person | 0.921 0.908 | 330 320 | 180 180 | |
| Painting of a person | Painting of a Disgusted person | 0.879 | 291 | 180 |
| Dataset | src_txt | trg_txt | XCLIP | Xrecon | from random noise |
| CelebA-HQ | Person | Young person | 0.8 | 0.905*3 | X |
| Face | Smiling face | 0.8 | 0.899*3 | X | |
| Face | Sad face | 0.8 | 0.894*3 | X | |
| Face | Angry face | 0.8 | 0.892*3 | X | |
| Face | Tanned face | 0.8 | 0.886*3 | X | |
| Face | Disgusted face | 0.8 | 0.880*3 | X | |
| Person | Person with makeup | 0.8 | 0.875*3 | X | |
| Person Photo | Person with curly hair | 0.8 | 0.835*3 | X | |
| Self-portrait by Frida Kahlo | 0.5 | 0.4433 | 0 | ||
| Human | Zombie | 0.6 | 0.868*3 | 0 | |
| Person | Nicolas Cage | 0.8 | 0.710*3 | 0 | |
| Human Photo | Painting in the style of Pixar | 0.8 | 0.667*3 | 0 | |
| Human | Painting in Modigliani style | 0.8 | 0.565*3 | 0 | |
| LSUN-church | Church | Neanderthal Gothic church | 1.2 | 0.802*3 | 0 |
| Church | Temple | 0.8 | 0.912*3 | 0 | |
| Church | Department store | 0.8 | 0.898*3 | 0 | |
| Church | Wooden house | 0.8 | 0.841*3 | 0 | |
| Church | Ancient traditional Asian tower | 0.8 0.8 | 0.793*3 | 0 | |
| Church | Red brick wall Church | 0.8 | 0.784*3 0.774*3 | 0 | |
| Church | Factory | 0.8 | 0.702*3 | 0 | |
| LSUN-bedroom | Bedroom | Princess bedroom | 1.5 | 0.912*3 | 0 0 |
| Bedroom | Hotel bedroom | 1.5 | 0.917*3 | ||
| AFHQ-dog | Dog | Happy dog | 0 | ||
| Dog | Sleepy dog | 1.5 | 0.883*3 | 0 | |
| Dog | 1.5 | 0.866*3 | 0 | ||
| Dog | Wolf Yorkshire terrier | 1.7 | 0.850*3 | 0 | |
| METFACES | Painting of a person | Painting of a Sad person | 2 1.5 | 0.690*3 0.921*3 | 0 X |
| Painting of a person | Painting of a Smiling person | 1.5 | 0.908*3 | X | |
| Painting of a person | Painting of a Disgusted person | 1.5 | 0.879*3 | X |
| CelebA-HQ-in-domain | CelebA-HQ-unseen-domain | |||||||||
| Smiling | Sad | Tanned | Pixar | Neanderthal | ||||||
| Sdir | SC | Sdir | SC | Sdir | SC | Sdir | SC | Sdir | SC | |
| Asyrp (ours) | 0.921 | 89.02% | 0.964 | 88.90% | 0.991 | 85.71% | 0.956 | 76.51% | 0.805 | 79.03 |
| DiffusionCLIP | 0.813 | 91.41% | 0.760 | 89.93% | 0.888 | 92.85% | 0.811 | 89.91% | 0.661 | 81.23% |
| LSUN_church | ||||||
| Department store | ancient | red brick | ||||
| Sdir | SC | Sdir | SC | Sdir | SC | |
| Asyrp (ours) | 0.778 | 57.62% | 0.943 | 62.65% | 0.989 | 65.83% |
| DIffusionCLIP | 0.661 | 54.50% | 0.907 | 64.82% | 0.964 | 65.02% |
| Dataset | Environment | BC | CQL | IQL | DT | TT | MOReL | Diffuser | DD |
| Med-Expert | HalfCheetah | 55.2 | 91.6 | 86.7 | 86.8 | 95 | 53.3 | 79.8 | 90.6 ±1.3 |
| Med-Expert | Hopper | 52.5 | 105.4 | 91.5 | 107.6 | 110.0 | 108.7 | 107.2 | 111.8 ±1.8 |
| Med-Expert | Walker2d | 107.5 | 108.8 | 109.6 | 108.1 | 101.9 | 95.6 | 108.4 | 108.8 ±1.7 |
| Medium | HalfCheetah | 42.6 | 44.0 | 47.4 | 42.6 | 46.9 | 42.1 | 44.2 | 49.1 ±1.0 |
| Medium | Hopper | 52.9 | 58.5 | 66.3 | 67.6 | 61.1 | 95.4 | 58.5 | 79.3 ±3.6 |
| Medium | Walker2d | 75.3 | 72.5 | 78.3 | 74.0 | 79 | 77.8 | 79.7 | 82.5±1.4 |
| Med-Replay | HalfCheetah | 36.6 | 45.5 | 44.2 | 36.6 | 41.9 | 40.2 | 42.2 | 39.3 ±4.1 |
| Med-Replay | Hopper | 18.1 | 95 | 94.7 | 82.7 | 91.5 | 93.6 | 96.8 | 100±0.7 |
| Med-Replay | Walker2d | 26.0 | 77.2 | 73.9 | 66.6 | 82.6 | 49.8 | 61.2 | 75±4.3 |
| Average | 51.9 | 77.6 | 77 | 74.7 | 78.9 | 72.9 | 75.3 | 81.8 | |
| Mixed | Kitchen | 51.5 | 52.4 | 51 | 1 | 、 | 1 | 1 | 65 ±2.8 |
| Partial | Kitchen | 38 | 50.1 | 46.3 | - | - | = | = | 57 ±2.5 |
| Average | 44.8 | 51.2 | 48.7 | - | 1 | - | - | 61 | |
| Hopper-* | Diffuser CondDiffuser CondMLPDiffuser Decision Diffuser | |||
| Med-Expert | 107.6 | 111.3 | 105.6 | 111.8 ±1.6 |
| Medium | 58.5 | 66.3 | 54.1 | 79.3 ±3.6 |
| Med-Replay | 96.8 | 76.5 | 66.5 | 100 ±0.7 |
| Environment | Diffuser | DD |
| Single Constraint - Stacking | 45.6 ±3.1 | 58.0 ±3.1 |
| Single Constraint - Rearrangement | 58.9 ±3.4 | 62.7 ±3.1 |
| Single Constraint Average | 52.3 | 60.4 |
| Multiple Constraints - Stacking | 1 | 60.3 ±3.1 |
| Multiple Constraints - Rearrangement | 67.2 ±3.1 | |
| Multiple Constraints Average | 1 | 63.8 |
| Condition | Trott | Pace | Bound |
| Only Bound | 0.8 | 1.0 | 98.2 |
| Only Pace | 1.4 | 97.7 | 0.9 |
| Bound + Pace | 1.4 | 38.5 | 60.1 |
| Decision Diffuser | Hopper-Medium-Expert |
| α=0 | 104.3 ± 0.7 |
| α= 0.5 | 111.8 ±1.6 |
| α= 1.0 | 107.1 ± 3.5 |
| Environment | BC | CondDiffuser | Decision Diffuser |
| Position Control | 57.3 ±1.2 | 87.3 ±3.1 | 87.8 ±2.8 |
| Torque Control | 55.2 ±1.5 | 71.8 ±3.4 | 84.7 ±2.2 |
| p | BC | Decision Diffuser | Diffuser | CQL |
| 0.00 | 55.2±1.5 | 84.7±2.2 | 72.4±1.4 | 73.2±2.3 |
| 0.05 | 49.3±3.6 | 77.3±3.1 | 63.2±2.9 | 61.8±3.7 |
| 0.10 | 25.8±3.8 | 53.2±4.1 | 52.3±4.6 | 51.2±4.3 |
| 0.15 | 15.1±4.3 | 41.3±4.9 | 41.6±5.1 | 42.2±5.5 |
| Dataset | Environment | BC | CQL | IQL | DT | TT | MOReL | Diffuser | DD |
| Med-Expert | HalfCheetah | 55.2 | 91.6 | 86.7 | 86.8 | 95 | 53.3 | 79.8 | 90.6 ±1.3 |
| Med-Expert | Hopper | 52.5 | 105.4 | 91.5 | 107.6 | 110.0 | 108.7 | 107.2 | 111.8 ±1.8 |
| Med-Expert | Walker2d | 107.5 | 108.8 | 109.6 | 108.1 | 101.9 | 95.6 | 108.4 | 108.8 ±1.7 |
| Medium | HalfCheetah | 42.6 | 44.0 | 47.4 | 42.6 | 46.9 | 42.1 | 44.2 | 49.1 ±1.0 |
| Medium | Hopper | 52.9 | 58.5 | 66.3 | 67.6 | 61.1 | 95.4 | 58.5 | 79.3 ±3.6 |
| Medium | Walker2d | 75.3 | 72.5 | 78.3 | 74.0 | 79 | 77.8 | 79.7 | 82.5±1.4 |
| Med-Replay | HalfCheetah | 36.6 | 45.5 | 44.2 | 36.6 | 41.9 | 40.2 | 42.2 | 39.3 ±4.1 |
| Med-Replay | Hopper | 18.1 | 95 | 94.7 | 82.7 | 91.5 | 93.6 | 96.8 | 100±0.7 |
| Med-Replay | Walker2d | 26.0 | 77.2 | 73.9 | 66.6 | 82.6 | 49.8 | 61.2 | 75±4.3 |
| Average | 51.9 | 77.6 | 77 | 74.7 | 78.9 | 72.9 | 75.3 | 81.8 | |
| Mixed | Kitchen | 51.5 | 52.4 | 51 | 1 | 、 | 1 | 1 | 65 ±2.8 |
| Partial | Kitchen | 38 | 50.1 | 46.3 | - | - | = | = | 57 ±2.5 |
| Average | 44.8 | 51.2 | 48.7 | - | 1 | - | - | 61 |
| Hopper-* | Diffuser CondDiffuser CondMLPDiffuser Decision Diffuser | |||
| Med-Expert | 107.6 | 111.3 | 105.6 | 111.8 ±1.6 |
| Medium | 58.5 | 66.3 | 54.1 | 79.3 ±3.6 |
| Med-Replay | 96.8 | 76.5 | 66.5 | 100 ±0.7 |
| Environment | Diffuser | DD |
| Single Constraint - Stacking | 45.6 ±3.1 | 58.0 ±3.1 |
| Single Constraint - Rearrangement | 58.9 ±3.4 | 62.7 ±3.1 |
| Single Constraint Average | 52.3 | 60.4 |
| Multiple Constraints - Stacking | 1 | 60.3 ±3.1 |
| Multiple Constraints - Rearrangement | 67.2 ±3.1 | |
| Multiple Constraints Average | 1 | 63.8 |
| Condition | Trott | Pace | Bound |
| Only Bound | 0.8 | 1.0 | 98.2 |
| Only Pace | 1.4 | 97.7 | 0.9 |
| Bound + Pace | 1.4 | 38.5 | 60.1 |
| Decision Diffuser | Hopper-Medium-Expert |
| α=0 | 104.3 ± 0.7 |
| α= 0.5 | 111.8 ±1.6 |
| α= 1.0 | 107.1 ± 3.5 |
| Environment | BC | CondDiffuser | Decision Diffuser |
| Position Control | 57.3 ±1.2 | 87.3 ±3.1 | 87.8 ±2.8 |
| Torque Control | 55.2 ±1.5 | 71.8 ±3.4 | 84.7 ±2.2 |
| p | BC | Decision Diffuser | Diffuser | CQL |
| 0.00 | 55.2±1.5 | 84.7±2.2 | 72.4±1.4 | 73.2±2.3 |
| 0.05 | 49.3±3.6 | 77.3±3.1 | 63.2±2.9 | 61.8±3.7 |
| 0.10 | 25.8±3.8 | 53.2±4.1 | 52.3±4.6 | 51.2±4.3 |
| 0.15 | 15.1±4.3 | 41.3±4.9 | 41.6±5.1 | 42.2±5.5 |
| Dataset | Environment | BC | CQL | IQL | DT | TT | MOReL | Diffuser | DD |
| Med-Expert | HalfCheetah | 55.2 | 91.6 | 86.7 | 86.8 | 95 | 53.3 | 79.8 | 90.6 ±1.3 |
| Med-Expert | Hopper | 52.5 | 105.4 | 91.5 | 107.6 | 110.0 | 108.7 | 107.2 | 111.8 ±1.8 |
| Med-Expert | Walker2d | 107.5 | 108.8 | 109.6 | 108.1 | 101.9 | 95.6 | 108.4 | 108.8 ±1.7 |
| Medium | HalfCheetah | 42.6 | 44.0 | 47.4 | 42.6 | 46.9 | 42.1 | 44.2 | 49.1 ±1.0 |
| Medium | Hopper | 52.9 | 58.5 | 66.3 | 67.6 | 61.1 | 95.4 | 58.5 | 79.3 ±3.6 |
| Medium | Walker2d | 75.3 | 72.5 | 78.3 | 74.0 | 79 | 77.8 | 79.7 | 82.5±1.4 |
| Med-Replay | HalfCheetah | 36.6 | 45.5 | 44.2 | 36.6 | 41.9 | 40.2 | 42.2 | 39.3 ±4.1 |
| Med-Replay | Hopper | 18.1 | 95 | 94.7 | 82.7 | 91.5 | 93.6 | 96.8 | 100±0.7 |
| Med-Replay | Walker2d | 26.0 | 77.2 | 73.9 | 66.6 | 82.6 | 49.8 | 61.2 | 75±4.3 |
| Average | 51.9 | 77.6 | 77 | 74.7 | 78.9 | 72.9 | 75.3 | 81.8 | |
| Mixed | Kitchen | 51.5 | 52.4 | 51 | 1 | 、 | 1 | 1 | 65 ±2.8 |
| Partial | Kitchen | 38 | 50.1 | 46.3 | - | - | = | = | 57 ±2.5 |
| Average | 44.8 | 51.2 | 48.7 | - | 1 | - | - | 61 |
| Hopper-* | Diffuser CondDiffuser CondMLPDiffuser Decision Diffuser | |||
| Med-Expert | 107.6 | 111.3 | 105.6 | 111.8 ±1.6 |
| Medium | 58.5 | 66.3 | 54.1 | 79.3 ±3.6 |
| Med-Replay | 96.8 | 76.5 | 66.5 | 100 ±0.7 |
| Condition | Trott | Pace | Bound |
| Only Bound | 0.8 | 1.0 | 98.2 |
| Only Pace | 1.4 | 97.7 | 0.9 |
| Bound + Pace | 1.4 | 38.5 | 60.1 |
| Environment | Diffuser | DD |
| Single Constraint - Stacking | 45.6 ±3.1 | 58.0 ±3.1 |
| Single Constraint - Rearrangement | 58.9 ±3.4 | 62.7 ±3.1 |
| Single Constraint Average | 52.3 | 60.4 |
| Multiple Constraints - Stacking | 1 | 60.3 ±3.1 |
| Multiple Constraints - Rearrangement | 67.2 ±3.1 | |
| Multiple Constraints Average | 1 | 63.8 |
| Decision Diffuser | Hopper-Medium-Expert |
| α=0 | 104.3 ± 0.7 |
| α= 0.5 | 111.8 ±1.6 |
| α= 1.0 | 107.1 ± 3.5 |
| Environment | BC | CondDiffuser | Decision Diffuser |
| Position Control | 57.3 ±1.2 | 87.3 ±3.1 | 87.8 ±2.8 |
| Torque Control | 55.2 ±1.5 | 71.8 ±3.4 | 84.7 ±2.2 |
| p | BC | Decision Diffuser | Diffuser | CQL |
| 0.00 | 55.2±1.5 | 84.7±2.2 | 72.4±1.4 | 73.2±2.3 |
| 0.05 | 49.3±3.6 | 77.3±3.1 | 63.2±2.9 | 61.8±3.7 |
| 0.10 | 25.8±3.8 | 53.2±4.1 | 52.3±4.6 | 51.2±4.3 |
| 0.15 | 15.1±4.3 | 41.3±4.9 | 41.6±5.1 | 42.2±5.5 |
| Condition | Trott | Pace | Bound |
| Only Trott | 97.2 | 1.7 | 1.1 |
| Only Pace | 0.8 | 98.1 | 1.1 |
| Trott + Pace | 55.6 | 43.6 | 0.8 |
| Condition | Trott | Pace | Bound |
| Only Trott | 96.4 | 2.2 | 1.4 |
| Only Bound | 1.6 | 0.6 | 97.8 |
| Trott + Bound | 51.3 | 0.9 | 47.8 |
| Condition | Trott | Pace | Bound |
| Only Bound | 0.8 | 1.0 | 98.2 |
| Only Pace | 1.4 | 97.7 | 0.9 |
| Bound + Pace | 1.4 | 38.5 | 60.1 |
| Parameter | Value |
| Uthro | -52 (mV) |
| Urest,Ureset | -65 (mV) |
| go | 0.05 (mV) |
| Tg | 10 (ms) |
| △tref | 5(ms) |
| Tm | 20(ms) |
| fmax | 128(Hz) |
| hin，Win | 22 |
| nout | [100,500,1000] |
| chlc | [25,50,100,250] |
| k | [11,13, 15,17] |
| S | [2, 3,4] |
| Tadapt,Tdec,Tlearn | 256 (ms) |
| (Npre,Npost)STDP | (0.0001,0.01) |
| (npre,post)R-STDP | (0.1,0.1) |
| 2 | 1 |
| nrpe | [(static),0.075,0.125,0.175,0.25] |
| a | 0.9 |
| Winh | -100 |
| Cnorm | 0.25 |
| Parameter | Value |
| Uthro | -52 (mV) |
| Urest,Ureset | -65 (mV) |
| go | 0.05 (mV) |
| Tg | 10 (ms) |
| △tref | 5(ms) |
| Tm | 20(ms) |
| fmax | 128(Hz) |
| hin，Win | 22 |
| nout | [100,500,1000] |
| chlc | [25,50,100,250] |
| k | [11,13, 15,17] |
| S | [2, 3,4] |
| Tadapt,Tdec,Tlearn | 256 (ms) |
| (Npre,Npost)STDP | (0.0001,0.01) |
| (npre,post)R-STDP | (0.1,0.1) |
| 2 | 1 |
| nrpe | [(static),0.075,0.125,0.175,0.25] |
| a | 0.9 |
| Winh | -100 |
| Cnorm | 0.25 |
| Parameters[k,s,nrpe,nout] | nneurons | nsynapses | Test accuracy | SVM test accuracy |
| [13,3,0.025,100] | 1700 | 430400 | 61.30 ±3.14 | 87.5±1.32 |
| [15,4,0.175,1000] | 1884 | 490000 | 75.00 ±2.68 | 83.3±1.74 |
| [15,4,0.125,1000] | 1884 | 490000 | 76.40 ±2.43 | 83.3±1.74 |
| [15,4,(static),100] | 984 | 130000 | 68.8±2.87 | 83.3±1.74 |
| Paper | Encoding | Architecture | Bio-plausibility criteria | Acc. |
| BioLCNet (proposed, RL) | rate-based | Locally connected+Dense | STDP, RL, LC | 76.40 |
| BioLCNet (proposed,SVM) | rate-based | Locally connected | STDP, LC | 87.5 |
| Beyeler et al. (2013) | rate-based | Dense | STDP | 91.60 |
| Diehl& Cook (2015) | rate-based | Dense | STDP | 95.00 |
| Tavanaei &Maida (2015) | rate-based | Dense | STDP | 75.93 |
| Allred & Roy (2016) | rate-based | Dense | STDP | 86.59 |
| Kheradpisheh et al. (2018) | rank-order | Convolutional | STDP | 98.40 |
| Saunders et al. (2018) | rate-based | Convolutional | STDP | 84.23 |
| Lee et al. (2018) | rate-based | Convolutional | STDP | 91.1 |
| Mozafari et al. (2019) | rank-order | Convolutional | STDP, RL | 97.2 |
| Saunders et al. (2019) | rate-based | Locally connected | STDP, LC | 95.07 |
| Parameters[k,s,nrpe,nout] | nneurons | nsynapses | Test accuracy | SVM test accuracy |
| [13,3,0.025,100] | 1700 | 430400 | 61.30 ±3.14 | 87.5±1.32 |
| [15,4,0.175,1000] | 1884 | 490000 | 75.00 ±2.68 | 83.3±1.74 |
| [15,4,0.125,1000] | 1884 | 490000 | 76.40 ±2.43 | 83.3±1.74 |
| [15,4,(static),100] | 984 | 130000 | 68.8±2.87 | 83.3±1.74 |
| Paper | Encoding | Architecture | Bio-plausibility criteria | Acc. |
| BioLCNet (proposed, RL) | rate-based | Locally connected+Dense | STDP, RL, LC | 76.40 |
| BioLCNet (proposed,SVM) | rate-based | Locally connected | STDP, LC | 87.5 |
| Beyeler et al. (2013) | rate-based | Dense | STDP | 91.60 |
| Diehl& Cook (2015) | rate-based | Dense | STDP | 95.00 |
| Tavanaei &Maida (2015) | rate-based | Dense | STDP | 75.93 |
| Allred & Roy (2016) | rate-based | Dense | STDP | 86.59 |
| Kheradpisheh et al. (2018) | rank-order | Convolutional | STDP | 98.40 |
| Saunders et al. (2018) | rate-based | Convolutional | STDP | 84.23 |
| Lee et al. (2018) | rate-based | Convolutional | STDP | 91.1 |
| Mozafari et al. (2019) | rank-order | Convolutional | STDP, RL | 97.2 |
| Saunders et al. (2019) | rate-based | Locally connected | STDP, LC | 95.07 |
| Parameter | Value |
| Uthro | -52 (mV) |
| Urest,Ureset | -65 (mV) |
| go | 0.05 (mV) |
| Tg | 10 (ms) |
| △tref | 5(ms) |
| Tm | 20(ms) |
| fmax | 128(Hz) |
| hin，Win | 22 |
| nout | [100,500,1000] |
| chlc | [25,50,100,250] |
| k | [11,13, 15,17] |
| S | [2, 3,4] |
| Tadapt,Tdec,Tlearn | 256 (ms) |
| (Npre,Npost)STDP | (0.0001,0.01) |
| (npre,post)R-STDP | (0.1,0.1) |
| 2 | 1 |
| nrpe | [(static),0.075,0.125,0.175,0.25] |
| a | 0.9 |
| Winh | -100 |
| Cnorm | 0.25 |
| Paper | Encoding | Architecture | Bio-plausibility criteria | Acc. |
| BioLCNet (proposed, RL) | rate-based | Locally connected+Dense | STDP, RL, LC | 76.40 |
| BioLCNet (proposed,SVM) | rate-based | Locally connected | STDP, LC | 87.5 |
| Beyeler et al. (2013) | rate-based | Dense | STDP | 91.60 |
| Diehl& Cook (2015) | rate-based | Dense | STDP | 95.00 |
| Tavanaei &Maida (2015) | rate-based | Dense | STDP | 75.93 |
| Allred & Roy (2016) | rate-based | Dense | STDP | 86.59 |
| Kheradpisheh et al. (2018) | rank-order | Convolutional | STDP | 98.40 |
| Saunders et al. (2018) | rate-based | Convolutional | STDP | 84.23 |
| Lee et al. (2018) | rate-based | Convolutional | STDP | 91.1 |
| Mozafari et al. (2019) | rank-order | Convolutional | STDP, RL | 97.2 |
| Saunders et al. (2019) | rate-based | Locally connected | STDP, LC | 95.07 |
| Parameters[k,s,nrpe,nout] | nneurons | nsynapses | Test accuracy | SVM test accuracy |
| [13,3,0.025,100] | 1700 | 430400 | 61.30 ±3.14 | 87.5±1.32 |
| [15,4,0.175,1000] | 1884 | 490000 | 75.00 ±2.68 | 83.3±1.74 |
| [15,4,0.125,1000] | 1884 | 490000 | 76.40 ±2.43 | 83.3±1.74 |
| [15,4,(static),100] | 984 | 130000 | 68.8±2.87 | 83.3±1.74 |