new

Get trending papers in your email inbox once a day!

Get trending papers in your email inbox!

Daily Papers

byAK and the research community

Oct 7

Submitted by

Nothing2Say

Rethinking Cross-Tokenizer On-Policy Distillation: From Alignment Coverage to Supervision Reliability

·
7 authors

1

Submitted by

JohnZhan

DuoMatching: Joint-Marginal Distribution Matching for Few-Step Video Generation

ByteDance

Submitted by

tuidan

TRACE: Rollout-Guided Quantization-Aware Training for FP4 Reinforcement Learning of MoE Language Models

Qwen

1

Submitted by

Qing145

EVISKILL: Grounding Skill Evolution in Replayable Evidence

JLU-Deep

DEEP Group at Jilin University

Submitted by

danielhzlin

From Evidence to Action: How Tool-Using Agents Fail

NationalUniversityofSingapore

National University of Singapore

Submitted by

DongkiKim

AutoSciBench: Autonomous Benchmark Generation for Evaluating Scientific Agents

Genentech

1

Submitted by

lshig96

Taming VLAs under Robot Execution Errors: Self-Compensation and Stress Testing

POSTECH

Pohang University of Science and Technology

2

Submitted by

Benyou

HuatuoGPT-3: RL-Only Domain Adaptation from Base Models

CUHKSZ

Chinese University of Hong Kong, Shenzhen

Submitted by

Jam1e3

World Action Learning via Interaction-Centric Spectral Latent Guidance

The Hong Kong University of Science and Technology

1

Submitted by

ekinderman

UNREAL: Unifying Retrieval and Long-Context with a Single Model

nvidia

1

Submitted by

shizhuo2

MiniCorp: The Last Mile of the AI Agent Firm

microsoft

Submitted by

zgzxy001

Selection-Based Structured Reasoning: Toward Efficient Multimodal Search Agents

CarnegieMellonU

Carnegie Mellon University

Submitted by

enrico-protom

AGO AI Quality Gate: Evidence-First Release Decisions for Retrieval-Augmented Generation

·
6 authors

1

Submitted by

Karn3003

DiffGate: Difficulty-Gated Teacher Guidance for On-Policy Distillation

Snapchat

Submitted by

phanviethoang1512

Rationale-Guided Policy Optimization: Learning to Reason with Adaptive Rationale Scaffolding

newyorkuniversity

New York University

Submitted by

dzxagent

GUI-HARVEST: Self-Improving GUI Agents through Evidence-Driven Harness Evolution

The Chinese University of Hongkong,Shenzhen

Submitted by

taesiri

EmbodiedSmith: Scaling Embodied Data through Recursive Self-Improvement Flywheel in Simulation

·
16 authors

Submitted by

YihuaiHong

Making LLMs Say What They Think: Measuring and Improving CoT-Interpretability Alignment

newyorkuniversity

New York University

Submitted by

gunho1123

SlimWise: Decoupling Expert Pruning Across Prefill and Decode for Efficient MoE Serving

·
6 authors

2

Submitted by

IdanAchi

Adaptive Latent Capacity for World Models

Arm

1

Submitted by

HollowMan6

NeMo-DCR: Bit-Exact Delta-Compressed Refit for Scalable Agentic RL at Trillion-Parameter Scale

nvidia

Submitted by

sam121796

DiVeR: Decision-Critical Verifier Learning for VLA Test-Time Scaling

MicrosoftResearch

Microsoft Research

Submitted by

zyanzhe

Sherpa: Teaching LLMs to Teach Adaptively

SALT-NLP

Social And Language Technology Lab

Submitted by

omer11a

Learning to Read the Contextual Tokens in Diffusion Transformers

tau

Tel Aviv University

Submitted by

Sarim-Hash

AdvSim2Real : Training Web Agents Against Adaptive Prompt Injection in a Web World Model

MBZUAI

Mohamed Bin Zayed University of Artificial Intelligence

Submitted by

qiusizhan

Understanding and Enhancing Backdoor Persistency in LLM Agent Post-Training

UIUC-CS

University of Illinois at Urbana-Champaign

Submitted by

Kyudan

Hiding Tool Latency in On-Device Cascaded Voice Agent through Speculative Execution

Qualcomm-AI-Research

Qualcomm AI Research

Submitted by

antoineedy

DAEDALUS: Bootstrapping Agent Memory from Self-Generated Tasks

illuin

Illuin Technology

Submitted by

jason-moon

Harness-Aware Distillation for Small Language Model Agents

kaist-ai

Submitted by

chenfeng1271

HLA: Expressive Hybrid Linear Attention via Chunk-Wise Dynamic Mixing

Monash University

Submitted by

HaoyuHuang2

Towards In-Parameter Memory Augmentation for Large Language Models

Submitted by

taesiri

DistScene: Object-to-Scene Distillation for 3D Scene Generation

·
9 authors

Submitted by

xinpeng

OPD Before RL: Warm-Starting Rubric-Based RL with On-Policy Distillation

meta

Submitted by

yoomee-cho

Accent Analogy Guidance: More Speaker Similarity at Equal Accent in Cross-Lingual Voice Cloning

est-ai

Submitted by

jaehong31

Attacca: Goal-Directed Control under State Continuity for Long-Horizon Embodied Agents

nanyang-technological-university-singapore

Nanyang Technological University Singapore

Submitted by

yuuxia

Personal-Agent Mediated Recommendation with Cross-Platform User History

meta

Submitted by

taesiri

VeriFine: Scaling Verification for Self-Improvement in Embodied Reasoning

nvidia

Submitted by

taesiri

World Models' Last Exam in Physics

·
11 authors

Submitted by

domejiraphon

Building Rome from a Single Image

·
8 authors

Submitted by

zhouzhenghong-gt

ALIVE: Interaction-Aligned Object Insertion for First-Frame-Guided Video Editing

university-of-rochester

University of Rochester

Submitted by

AllenJin

Harness Engineering for Software Engineering via Modular Executable Dev-Primitives

·
4 authors

Submitted by

Kyudan

HiPLEX: Hierarchical Policy Factorization for Full Duplex Speech Language Models

Qualcomm-AI-Research

Qualcomm AI Research

Submitted by

mwbini

Learning Functional Subspaces for Neural Network Compression

·
8 authors

1

Submitted by

lucasbandarkar

Multilinguality in Hybrid Attention LLMs

UCLA

University of California, Los Angeles

1

Submitted by

Bochkov

Do Language Models Need a Trainable Input Embedding Table? Fixed Minimal Token Codes at 1.7B-Class Scale

·
1 authors

Submitted by

zhangdd24

Learning Discriminative Geometry for Drifting Models

alpintosh

Aalto University

Submitted by

shreshthsaini

JLD: Perceptual Distance Through A Jacobian Lens

·
3 authors

Submitted by

benni23

Judged Useless, Queried Anyway: Tool-Using Agents Rarely Turn Their Own Evidence Judgments into Stopping Decisions

nanyang-technological-university-singapore

Nanyang Technological University Singapore

Submitted by

arnav1993k

Stepped MoE: Segment-Level Routing with Configurable Inference Complexity

apple

Submitted by

tumeteor

A Safe Action Is Not Enough: Feasible-Future Decoding for Vision-Language-Action Policies

huawei-noah

HUAWEI Noah's Ark Lab

Submitted by

shreshthsaini

MEND: RL For Flow Models via Proximal Velocity Matching

UTEXAS

University of Texas at Austin

Submitted by

abdelghanibelgaid

Technical Report on the Turba Fertilizer Machine Learning Stack in Morocco

open-turba

Submitted by

yuzhe-yang

Sensor-Language-Action Models

yang-ai-lab

Health Intelligence Lab @ UCLA

Submitted by

Benchanything

CheckerBench: Can Long-Horizon Agents Synthesize Static-Analysis Checkers?

humanlaya-data-lab

Humanlaya Data Lab

Submitted by

jwfanDL

Kinematic MeanFlow: One-Step Action Generation Policy for Robotic Foundation Models

IntelLabsChina

Intel Labs China

Submitted by

lucasbandarkar

Cross-Lingual Alignment for Decoder-Only Models using MoE Routers

UCLA

University of California, Los Angeles

Submitted by

jefequien

S2PD: Serial-to-Parallel Diffusion for Physically and Logically Consistent Video Generation

CambUni

University of Cambridge

Submitted by

KevinZhangZJU

Conditional Trajectory Peaks: Single-Pass Multimodal Policies over Action Chunks

·
7 authors

1

Submitted by

KevinZhangZJU

Execution-Aligned Progressive Noise for Consistent Asynchronous Replanning in Generative Robot Policies

·
6 authors

1

Submitted by

KevinZhangZJU

Toward Real-Time VLAs: Stage-Aware Two-Step Flow Denoising and System-Level Evaluation

·
11 authors

Submitted by

KevinZhangZJU

Magic-W0: A Structured World-Action Foundation Model for Physical Intelligence

·
18 authors

Submitted by

turhancan97

WildMatch: Weakly Supervised Image Matcher Adaptation for Wildlife Re-Identification

·
6 authors

Submitted by

jarmstrong001

Source Identification Is Not Fitness Testing: Measuring the Limits of Synthetic-Data Attribution

·
1 authors

3

Submitted by

wrecka

CtrlCache: Accelerating Interactive Video World Models with Control-Aware Caching

·
5 authors

Submitted by

Yoniel

ConEx: Human-Interpretable Saliency Maps via Concept-Aware Attribution

·
4 authors

Submitted by

omi-n

JumpStart Your Policy Learning with Lessons from 160,000 Training Runs

raivn

RAIVN: Reasoning, AI, & Vision Lab @ UW

Submitted by

ZhengmingYu

DMAD: Distribution Matching as Adversarial Distillation for Fast Visual Generation

ByteDance

Submitted by

Aoraku

DeCoPrune: Efficient KV-Cache Pruning for Autoregressive Video Diffusion via Denoising Consistency

mmlab-ntu