Good Papers: read the good ones

Intern-Atlas builds a methodological evolution graph from over one million AI papers to model how methods emerge and adapt, enabling automated idea evaluation and generation.

Yujun Wu, Dongxu Zhang, Xinchen Li, Jinhang Xu and 10 more

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 47 on Hugging Face

– ReadersNo votes yet

11/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

71%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026ShanghaiTech Shanghai Jiao Tong The Chinese University of Hong K Shanghai Jiaotong Morphic Robotics Robot manipulation

Implicit Drifting Policy: One-Step Action Generation via Conditional Expert Geometry

IDP uses conditional expert geometry to implicitly enforce training-time drifting corrections, yielding a one-step generative policy that improves robot control speed while maintaining valid action manifolds.

Zemin Yang, Yaoyu He, Yiming Zhong, Yuhao Zhang and 4 more

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

7/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 7 of 20 reviewers recommend it

lenient 3/5

medium 4/10

strict 0/5

71%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Shanghai Jiaotong Locomotion

SAGAS: Semantic-Aware Graph-Assisted Stitching for Offline Temporal Logic Planning

SAGAS learns a reusable latent reachability graph from fixed offline trajectory fragments to synthesize cost-aware accepting plans for unseen linear temporal logic specifications via test-time semantic graph augmentation and Büchi search without online interaction or retraining.

Ruijia Liu, Ancheng Hou, Xiang Yin

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

7/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

86%Must read

?Must readVote to see the score

NeurIPS 2026Shanghai Jiaotong Academy of Military Science Huawei Technologies Ltd.Shanghai Jiao Tong Flow matching

Accelerating Rectified Flow Models via Trajectory-Aware Caching

TACache decomposes rectified flow velocity errors into magnitude and direction components to skip steps and reconstruct velocities without extra evaluations, achieving up to 4.14x faster image and 2.11x faster video generation.

Xiao Liu, Kai Liu, Naiyang Guan, Hongliang Lu and 4 more

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

14/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Fudan Shanghai Jiaotong nanjing university Peng Cheng Laboratory Shanghai Jiao Tong Agent memory

MemForest: Efficient Agent Memory Management via EventTree Partitioning and Progressive Merging

MemForest partitions agent memory into event trees and progressively merges redundant nodes to cut storage and retrieval costs while preserving nearly all performance.

JunXi Wang, Jiayi Zhu, Te Sun, Chen Zhang and 8 more

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

11/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

71%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Shanghai Jiao Tong Huawei Technologies Ltd.Shanghai Jiaotong 3D generation

WorldAct: Activating Monolithic 3D Worlds into Interactive-Ready Object-Centric Scenes

WorldAct converts static generated 3D worlds into editable, interaction-ready scenes via multimodal decomposition and object reconstruction to enable manipulation and embodied tasks.

Jichen Hu, Jiawei Guo, Jiazhong Cen, Chen Yang and 2 more

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 5 on Hugging Face · Code ★ 32

– ReadersNo votes yet

7/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

83%Must read

?Must readVote to see the score

NeurIPS 2026Shanghai Jiao Tong SJTU Shanghai Jiaotong WeChat, Tencent Tencent AI Platform Quantization

D$^2$Quant: Accurate Low-bit Post-Training Weight Quantization for LLMs

D²Quant improves sub-4-bit LLM weight-only quantization via dual-scale quantizers for down-projection matrices and deviation-aware LayerNorm correction, boosting accuracy without extra bit budget.

Xianglong Yan, chengzhu bao, Zhiteng Li, Tianao Zhang and 4 more

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 1 on Hugging Face

– ReadersNo votes yet

13/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

88%Must read

?Must readVote to see the score

NeurIPS 2026SpotlightUC San Diego Shanghai Jiaotong U Science and Technology of Chin JD.com Zhejiang Video-language models

AdaCodec: A Predictive Visual Code for Video MLLMs

AdaCodec uses predictive visual codes to send full reference frames only when unpredictable, cutting video MLLM tokens by 7x while improving long-video benchmark scores and reducing latency.

Haowen Hou, Zhen Huang, Zheming Liang, Qingyi Si and 7 more

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 7 on Hugging Face

– ReadersNo votes yet

15/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

71%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Shanghai Jiaotong Tsinghua Hong Kong Polytechnic U Science and Technology of Chin Microsoft Robot manipulation

DeformGen: Dynamics-Based Topology Augmentation for Deformable Manipulation Policy Learning

DeformGen uses dynamics-based topological augmentation to generate diverse deformable object states and warp trajectories for improved manipulation policy learning.

Zili Lin, Wenyao Zhang, Yuyang Zhang, Zekun Qi and 8 more

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

7/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

83%Must read

?Must readVote to see the score

NeurIPS 2026SUN YAT-SEN UNIVERSITY South China University of Techno Harbin Institute of Technology Institute of automation, Chinese Shanghai Jiaotong Vision-language-action models

RePO-VLA: Recovery-Driven Policy Optimization for Vision-Language-Action Models

RePO-VLA improves vision-language-action robustness by assigning roles to success, recovery, and failure trajectories, raising adversarial success from 20% to 75%.

Weijia Liufu, Xiaoyu Guo, Ruiyi Chen, Jingzhi Liu and 15 more

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

13/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

83%Must read

?Must readVote to see the score

NeurIPS 2026Beijing Jiaotong Hong Kong Baptist Shanghai Jiaotong Nanyang Technological Video-language models

VideoOdyssey: A Benchmark for Ultra-Long-Context and Omni-Modal Video Understanding

VideoOdyssey benchmarks ultra-long video understanding via continuous certificates averaging 16 minutes, revealing MLLM failures in continuous reasoning and omni-modal perception.

He Haichen, Jiayi Zhou, Sifeng SHANG, Yihan Hu and 2 more

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 4 on Hugging Face

– ReadersNo votes yet

13/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

80%Must read

?Must readVote to see the score

NeurIPS 2026The University of Hong Kong Fudan Shanghai Jiao Tong Shanghai Jiaotong Tsinghua Robot manipulation

HandEdit: A Unified Benchmark for Egocentric Human-to-Robot Dexterous Hand Image Editing

HandEdit provides a 200M-instance benchmark and dataset for transforming egocentric human hands into diverse dexterous robot embodiments via image editing. It evaluates 11 baselines across hand-only and hand-arm tracks with embodiment-aware metrics.

Zhenjie Yang, Xingyu Jiao, Guopeng Zhong, Shuzhe Yang and 17 more

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

12/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

80%Must read

?Must readVote to see the score

NeurIPS 2026Fudan Shanghai Jiaotong U Science and Technology of Chin Huazhong University of Science a Shanghai Artificial Intelligence Agent memory

Experience Makes Skillful: Enabling Generalizable Medical Agent Reasoning via Self-Evolving Skill Memory

SkeMex improves medical agents via self-evolving skill memory that distills reusable procedural knowledge, governs retention by utility, and outperforms memory-based agents across clinical tasks.

Haoran Sun, Wenjie Li, Yujie Zhang, Zekai Lin and 7 more

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 4 on Hugging Face

– ReadersNo votes yet

12/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

86%Must read

?Must readVote to see the score

NeurIPS 2026The Chinese University of Hong K Shanghai Jiaotong SenseTime Reseach Spoken language models

MultiTalk: Scaling Full-Duplex Speech Models to Long, Multi-Party, Bilingual Conversation

MultiTalk introduces 57.6k hours of synthetic multi-party bilingual dialogue data and MultiTalkBench for long-form full-duplex evaluation, training a model that sustains coherent extended multi-party English-Chinese conversation and outperforms open-source baselines.

Ke Wang, Houxing Ren, Zimu Lu, Yunqiao Yang and 3 more

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

14/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 14 of 20 reviewers recommend it

lenient 5/5

medium 7/10

strict 2/5

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026SpotlightNankai Monash Shanghai Jiaotong Fudan Shanghai Jiao Tong LLM evaluation & benchmarks

IndustryCode: A Benchmark for Industry Code Generation

IndustryCode is a multi-domain, multi-language benchmark of 579 industrial coding sub-problems; Claude 3.5 Opus reaches 68.1% sub-problem and 42.5% main-problem accuracy.

Puyu Zeng, Zhaoxi Wang, Zhixu Duan, Liang Feng and 7 more

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

10/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

88%Must read

?Must readVote to see the score

NeurIPS 2026Shanghai Jiaotong Tencent Huawei Technologies Ltd.The Hong Kong University of Scie U Edinburgh, University of Edinb Hallucination & factuality

BALTO: Balanced Token-Level Policy Optimization for Hallucination Mitigation

BALTO applies balanced token-level credit assignment to mitigate LLM hallucinations by redistributing probability from unsupported to faithful content, outperforming response-level methods on faithfulness benchmarks.

Ning Li, Zixuan Guo, Yan Xu, Wenbo Fei and 6 more

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

15/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

72%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Shanghai Jiao Tong Shanghai Jiaotong Robot manipulation

PO-PDDL: Learning Symbolic POMDPs from Visual Demonstrations for Robot Planning Under Uncertainty

PO-PDDL learns symbolic POMDPs from robot videos to enable belief-space planning under partial observability and stochasticity, outperforming prior methods with lower planning cost.

Wenjing Tang, Jin Xuanjin, Yuan Liu, RenMing Huang and 2 more

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

8/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

83%Must read

?Must readVote to see the score

NeurIPS 2026The Hong Kong University of Scie The Hong Kong University of Scie Bournemouth Shanghai Jiaotong Beijing University of Posts and Vision-language-action models

Beyond Appearance Shifts: Task-Semantic Action Calibration for VLA Models

BAS-VLA calibrates frozen VLA actions via breaking-centered calibration and selective preservation gating to suppress stale-task drift and separate semantics. It achieves 98% clean success, 0% under target swaps, and 70% under style shifts versus 42%.

Shuaijun Liu, Feiyang You, Chengyu Wu, Shuyang Hao and 5 more

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

13/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

86%Must read

?Must readVote to see the score

NeurIPS 2026The Chinese University of Hong K Shanghai Jiaotong national university of singaore,Advanced Machine Intelligence ByteDance Inc.Autoregressive generation

BitDance: Scaling Autoregressive Generative Models with Binary Tokens

BitDance is an autoregressive image generator that predicts binary visual tokens via a diffusion head and next-patch decoding, achieving state-of-the-art FID with far fewer parameters and much faster inference.

Yuang Ai, Jiaming Han, Shaobin Zhuang, Weijia Mao and 7 more

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 46 on Hugging Face · Code ★ 485

– ReadersNo votes yet

14/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

70%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026East China Normal Fudan Shanghai Jiaotong U Science and Technology of Chin Ant Group RL for LLMs

DiPO: Disentangled Perplexity Policy Optimization for Fine-grained Exploration-Exploitation Trade-Off

DiPO disentangles perplexity into exploration and exploitation subspaces to enable fine-grained trade-offs, improving LLM reasoning and function calling via stable perplexity-guided policy optimization.

Xiaofan Li, Ming Yang, Zhiyuan Ma, Shichao Ma and 8 more

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 13 on Hugging Face

– ReadersNo votes yet

4/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

86%Must read

?Must readVote to see the score

NeurIPS 2026Shanghai Jiao Tong Xiaohongshu, Inc.Ant Group Shanghai Jiaotong U Science and Technology of Chin LLM pretraining & scaling laws

xHC: Expanded Hyper-Connections

xHC expands Transformer hyper-connections beyond four streams via sparse updates and temporal augmentation, improving scaling efficiency. It boosts 18B MoE downstream scores by 4.0 points over mHC with lower compute and reduced memory traffic via xHC-Flash.

Xiangdong Zhang, Xiaohan Qin, Tuo Dai, Xiaoming Shi and 7 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 54 on Hugging Face · Code ★ 68

– ReadersNo votes yet

14/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 14 of 20 reviewers recommend it

lenient 3/5

medium 8/10

strict 3/5

86%Must read

?Must readVote to see the score

NeurIPS 2026U Electronic Science and Technol Xiaohongshu U the Chinese Academy of Science Shanghai Jiaotong National University of Singapore LLM reasoning & chain of thought

Hint Tuning: Less Data Makes Better Reasoners

Hint Tuning calibrates reasoning depth by using an instruct model as a difficulty probe, cutting tokens by 24-66% with 1K samples.

Siqi Fan, Minghao Li, Xiaoqian Ma, Xiusheng Huang and 5 more

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

14/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 14 of 20 reviewers recommend it

lenient 4/5

medium 8/10

strict 2/5

71%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Shanghai Jiaotong Xiaohongshu Inc.Shanghai Jiao Tong Efficient inference & serving

SageSched: Efficient LLM Scheduling Confronting Demand Uncertainty and Hybridity

SageSched predicts LLM output-length distributions and schedules via compute-and-memory cost models, improving efficiency by over 28.7%.

Zhenghao Gan, Yichen Bao, Yifei Liu, Chen Chen and 2 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

7/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

80%Must read

?Must readVote to see the score

NeurIPS 2026SpotlightShanghai Jiaotong California Institute of Technolo Representation & contrastive learning

Winfree Oscillatory Neural Network

Winfree Oscillatory Neural Network applies generalized synchronization dynamics to vision and reasoning tasks, scaling to ImageNet-1K and achieving 80.1% Maze-hard accuracy with 1% of prior parameters.

Jiawen Dai, Yue Song

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 1 on Hugging Face

– ReadersNo votes yet

12/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

72%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Tianjin Montreal Institute for Learning Shanghai Jiaotong Huawei Technologies Ltd.Nanyang Technological Robot manipulation

ForceFlow: Learning to Feel and Act via Contact-Driven Flow Matching

ForceFlow uses force-aware flow matching with asymmetric multimodal fusion and vision-to-force handover to achieve robust contact-rich manipulation with 37% higher success and stronger zero-shot generalization.

Shuoheng Zhang, Yifu Yuan, Hongyao Tang, YAN ZHENG and 6 more

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

8/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

83%Must read

?Must readVote to see the score

NeurIPS 2026U California, Berkeley Massachusetts Institute of Techn Shanghai Jiaotong Tsinghua NVIDIA ML systems & compilers

Flash-KMeans: Fast and Memory-Efficient Exact K-Means

Flash-KMeans eliminates GPU HBM bottlenecks via fused assignment and inverse mapping updates, delivering up to 17.9x speedups over existing exact k-means implementations.

Shuo Yang, Haocheng Xi, Yilong Zhao, Muyang Li and 9 more

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026 · ▲ 85 on Hugging Face · Code ★ 735

– ReadersNo votes yet

13/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

88%Must read

?Must readVote to see the score

NeurIPS 2026Beijing Academy of Artificial In Shanghai Jiaotong U Electronic Science and Technol Hainan The Hong Kong University of Scie 3D vision & reconstruction

Focusable Monocular Depth Estimation

FocusDepth uses spatially-aligned multi-scale prompt fusion to boost target-region depth accuracy and sharp boundaries while preserving global geometry, outperforming global baselines on FDE-Bench.

Yuxin Du, Tao Lin, Zile Zhong, Runting Li and 6 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

15/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

70%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026The Hong Kong University of Scie ByteDance Inc.Bytedance Hong Kong University of Science Shanghai Jiaotong Video generation

HOMIE: Human-object Centric Video Personalization via Multimodal Intelligent Enhancement

HOMIE unifies inter- and intra-subject video personalization via multimodal guidance and reference embeddings, achieving state-of-the-art human-object interaction fidelity.

Yiyang Cai, Nan Chen, Rongchang Xie, Junwen Pan and 7 more

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 54 on Hugging Face · Code ★ 177

– ReadersNo votes yet

4/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 4 of 20 reviewers recommend it

lenient 1/5

medium 3/10

strict 0/5

89%Must read

?Must readVote to see the score

NeurIPS 2026SpotlightNanyang Technological Shanghai Jiaotong Microsoft Shanghai Jiao Tong College of Computing and Data Sc Pose & motion

The Surprising Effectiveness of Video Diffusion Models for Hand Motion Reconstruction

ViDiHand leverages pretrained video diffusion models to reconstruct 4D hand poses directly from full egocentric video without detectors, substantially outperforming prior methods on ARCTIC, HOT3D, and HOI4D.

Yuxi Wang, Chengkai Jin, Yufei Liu, Wenqi Ouyang and 5 more

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 9 on Hugging Face · Code ★ 187

– ReadersNo votes yet

16/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

80%Must read

?Must readVote to see the score

NeurIPS 2026Nanjing University of Science an U Washington, Seattle Shanghai Jiaotong Fudan Nanjing Model-based RL

Distill to Think, Foresee to Act: Cognitive-Physical Reinforcement Learning for Autonomous Driving

CoPhy distills vision-language cognition into a BEV encoder and pairs it with an auto-regressive world model for action-conditioned forecasting to enable reinforcement learning with dual physical and cognitive rewards, achieving state-of-the-art autonomous driving results.

Yang Wu, Qiang Meng, Zhaojiang Liu, Youquan Liu and 2 more

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

12/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

88%Must read

?Must readVote to see the score

NeurIPS 2026Imperial College London Tsinghua Shanghai Jiaotong Unified multimodal models

Model Guides You How to Draw: Adaptive Visual Gating for Unified Multimodal Reasoning

AdaViG uses internal generation-intent and visual-fidelity signals to abort unhelpful visual reasoning steps early, improving multimodal reasoning accuracy by up to 5.7% while cutting visual generation costs by 25-91%.

Wenxi Gao, Guanxi Lu, Didi Zhu, Hao Chen and 4 more

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

15/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

86%Must read

?Must readVote to see the score

NeurIPS 2026The University of Hong Kong Shanghai Jiaotong Fudan Beijing University of Aeronautic The Chinese University of Hong K Embodied agents

Thinking with Imagination: Agentic Visual Spatial Reasoning with World Simulators

Astra enhances vision-language model spatial reasoning by letting agents generate imagined simulator views via RL, improving MMSI-Bench scores over direct answering.

Chenming Zhu, Jingli Lin, Yilin Long, Peizhou Cao and 3 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 16 on Hugging Face · Code ★ 32

– ReadersNo votes yet

14/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

80%Must read

?Must readVote to see the score

NeurIPS 2026Shanghai Jiao Tong Shanghai Jiaotong Text-to-image

DynT2I-Eval: A Dynamic Evaluation Framework for Text-to-Image Models

DynT2I-Eval dynamically generates fresh prompts across semantic dimensions to evaluate text-to-image models with continuously updated pairwise rankings.

Wang Juntong, Wang Jiarui, Huiyu Duan, Lewei Li and 2 more

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

12/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

86%Must read

?Must readVote to see the score

NeurIPS 2026Shanghai Jiaotong Eastern Institute of Technology,Efficient inference & serving

Adaptive Mass-Segmented KV Compression for Long-Context Reasoning

AMS replaces global token eviction with adaptive region-aware KV quotas to prevent reasoning block wipe-out, boosting long-context performance without extra attention overhead.

Junzhe Yang, Xiaoyu Shen

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026

– ReadersNo votes yet

14/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

83%Must read

?Must readVote to see the score

NeurIPS 2026Shanghai Jiaotong U Science and Technology of Chin ByteDance Inc.Eastern Institute of Technology,Text-to-image

Generation Navigator: A State-Aware Agentic Framework for Image Generation

Generation Navigator is a state-aware multi-turn text-to-image agent that learns to steer generation via trajectory-level reinforcement learning, achieving a 0.90 WISE score and 79.06% reasoning accuracy.

Jinming Liu, Ruoyu Feng, Yuqi Wang, Wenjun Zeng and 1 more

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

13/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

86%Must read

?Must readVote to see the score

NeurIPS 2026Shanghai Jiao Tong Shanghai Jiaotong Multimodal reasoning

Reflection Anchors for Propagation-Aware Visual Retention in Long-Chain Multimodal Reasoning

RAPO uses information-theoretic lower bounds to select high-entropy reflection anchors and optimize visual dependence via GRPO, substantially improving long-chain multimodal reasoning with reduced visual fading.

Xuan Gong, HanBo Huang, Hao Zheng, Yiran Zhang and 3 more

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

14/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026The University of Sydney ByteDance Inc.TikTok U Sydney Shanghai Jiaotong Video-language models

From Failure Taxonomy to Intervention: A Diagnostic Methodology for Industry-Scale AVLM in Video and Live-Streaming Platform Moderation

A diagnostic taxonomy maps AVLM failure signatures to targeted development interventions, enabling traceable industry-scale video moderation system improvements.

Shuchang Ye, Jinqiang Yu, Zhujun Xiao, Yajing Kong and 5 more

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

11/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

88%Must read

?Must readVote to see the score

NeurIPS 2026Wuhan Fudan Zhejiang Shanghai Jiaotong JD Explore Academy RL for LLMs

Visual-ERM: Reward Modeling for Visual Equivalence

Visual-ERM is a multimodal generative reward model that evaluates vision-to-code outputs in rendered visual space, improving Qwen3-VL-8B-Instruct by up to 8.4 points and outperforming larger models on fine-grained visual discrepancy benchmarks.

Ziyu Liu, Shengyuan Ding, Xinyu Fang, Xuanlang Dai and 5 more

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 21 on Hugging Face · Code ★ 69

– ReadersNo votes yet

15/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 15 of 20 reviewers recommend it

lenient 5/5

medium 8/10

strict 2/5

70%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Shanghai Jiao Tong Shanghai Jiaotong Xiaohongshu Inc.Efficient inference & serving

Justitia: Fair and Efficient Scheduling of Task-parallel LLM Agents with Selective Pampering

Justitia schedules task-parallel LLM agents via memory-centric cost prediction and virtual-time fair queuing to improve efficiency while preserving fairness and worst-case delays.

Mingyan Yang, Guanjie Wang, Manqi Luo, Yifei Liu and 5 more

Atlanta Poster Session 3, Thu, Dec 10, 10:00 AM–1:00 PM, Hall C1 · Published 2026

– ReadersNo votes yet

5/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 5 of 20 reviewers recommend it

lenient 4/5

medium 1/10

strict 0/5

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026SUN YAT-SEN UNIVERSITY The Hong Kong University of Scie Shanghai Jiaotong Shanghai Jiao Tong Video-language models

V-CAST: Video Curvature-Aware Spatio-Temporal Pruning for Efficient Video Large Language Models

V-CAST prunes video tokens via curvature-guided temporal budgets and dual-anchor spatial selection, achieving 98.6% original performance with 86.4% latency.

Xinying Lin, Xuyang Liu, Yiyu Wang, Teng Ma and 3 more

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

11/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 11 of 20 reviewers recommend it

lenient 4/5

medium 7/10

strict 0/5

91%Must read

?Must readVote to see the score

NeurIPS 2026Shanghai Jiao Tong Shanghai Jiaotong Multimodal reasoning

DRScaffold: Boosting Dense-Scene Reasoning in Lightweight Vision Language Models

DRScaffold improves lightweight vision-language model reasoning via structured four-stage supervision, surpassing a frozen 32B model on DRBench.

Xinrui Shi, Kai Liu, Ziqing Zhang, Jianze Li and 2 more

Atlanta Poster Session 6, Fri, Dec 11, 4:30 PM–7:30 PM, Hall C1 · Published 2026

– ReadersNo votes yet

17/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 17 of 20 reviewers recommend it

lenient 5/5

medium 9/10

strict 3/5

80%Must read

?Must readVote to see the score

NeurIPS 2026Shanghai Jiao Tong University, S U Cambridge Shanghai Jiao Tong Shanghai Jiaotong Agibot Vision-language-action models

Continually Evolving Skill Knowledge in Vision Language Action Model

Stellar VLA learns evolving skills via parameter-free continual imitation with knowledge-guided routing, achieving strong LIBERO performance with 1% replay and real-world transfer.

Yuxuan Wu, Guangming Wang, Zhiheng Yang, Tianchen Deng and 3 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

12/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 12 of 20 reviewers recommend it

lenient 4/5

medium 8/10

strict 0/5

89%Must read

?Must readVote to see the score

NeurIPS 2026Shanghai Jiao Tong Tsinghua Tsinghua & Sharpa Shanghai Jiaotong Robot manipulation

EgoTac: In-the-wild Tactile Prediction from Egocentric Vision

EgoTac predicts tactile signals from egocentric videos using 5.7M image-tactile pairs, achieving under 0.06N force error and outperforming contact estimators.

Wenkang Zhang, Chengbo Yuan, Zicheng Zhang, Zhengxue Cheng and 1 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

16/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 16 of 20 reviewers recommend it

lenient 5/5

medium 8/10

strict 3/5

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Shenzhen Technology Shanghai Jiaotong Shanghai Polytechnic Shanghai Jiao Tong Reward models & LLM-as-a-judge

Teacher-Aware Evolution of Heuristic Programs from Learned Optimization Policies

A teacher-aware evolutionary framework uses learned optimization policies as behavioral teachers to evolve static executable heuristics, improving combinatorial optimization benchmarks without neural inference at deployment.

Minyu Chen, Song Qin, Ling-I Wu, Jianxin Xue and 1 more

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

11/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 11 of 20 reviewers recommend it

lenient 5/5

medium 5/10

strict 1/5

74%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Shanghai Jiaotong Fudan Nanjing Shanghai Jiao Tong Xiamen Vision-language models

ViCO: A Training Strategy towards Semantic Aware Dynamic High-Resolution

ViCO minimizes vision tokens via consistency training across compression ratios, cutting tokens up to 50% while preserving capabilities through semantic-based routing.

Long Cui, Weiyun Wang, Jie Shao, Zichen Wen and 4 more

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 5 on Hugging Face

– ReadersNo votes yet

9/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 9 of 20 reviewers recommend it

lenient 5/5

medium 4/10

strict 0/5

86%Must read

?Must readVote to see the score

WWW 2026Google DeepMind Shanghai Jiaotong Segmentation

GMOS: Grounding Moving Object Segmentation in 3D Space and Time

GMOS grounds moving object segmentation in 3D space and time using an RGB video framework, achieving state-of-the-art results across MOS benchmarks with faster online inference.

Junyu Xie, Tengda Han, Weidi Xie, Andrew Zisserman

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

14/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 14 of 20 reviewers recommend it

lenient 4/5

medium 8/10

strict 2/5

80%Must read

?Must readVote to see the score

NeurIPS 2026U Electronic Science and Technol Shanghai Jiaotong Beijing University of Posts and Chongqing Afari Intelligent Driv Megvii Technology Inc.Embodied agents

CoWorld-VLA: Thinking in a Multi-Expert World Model for Autonomous Driving

CoWorld-VLA embeds multi-expert world tokens into vision-language-action models and couples diffusion planning with scene context to generate continuous ego trajectories, improving autonomous driving performance.

Jingqi Wang, minqing huang, Zihan Liang, Yujiao Xiang and 6 more

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

12/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 12 of 20 reviewers recommend it

lenient 5/5

medium 6/10

strict 1/5

86%Must read

?Must readVote to see the score

NeurIPS 2026Shanghai Jiao Tong Shanghai Jiaotong Huawei Technologies Ltd.Unified multimodal models

G$^2$TR: Generation-Guided Visual Token Reduction for Separate-Encoder Unified Multimodal Models

G²TR uses generation-branch signals to reduce visual tokens in unified multimodal models, cutting prefill computation by 1.94× while preserving reasoning and editing performance.

Junxian Li, Kai Liu, Zizhong Ding, Zhixin Wang and 3 more

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

14/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 14 of 20 reviewers recommend it

lenient 5/5

medium 9/10

strict 0/5

80%Must read

?Must readVote to see the score

NeurIPS 2026Jilin Shanghai Jiao Tong ByteDance Inc.The University of Sydney U Sydney Vision-language models

EasyLens: A Training-Free Plug-and-Play Subtle-Lesion Representation Amplifier for Medical Vision-Language Models

EasyLens is a training-free plug-and-play module that amplifies subtle lesion representations in frozen medical vision-language models via prototype-based patch selection and morphology-guided residual enhancement, improving detection across datasets.

QIWEI ZENG, Hao Wang, Jinghao Lin, Shuchang Ye and 5 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

12/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 12 of 20 reviewers recommend it

lenient 5/5

medium 7/10

strict 0/5

Show 20 more papers