Good Papers: read the good ones

Event Cascade Pruning uses event-camera motion cues to prune video tokens in first-person spatial reasoning, improving accuracy by 1.31 points with 80% fewer tokens and 1.89x speedup.

Pengtao Ma, Ziliang Zhou, Ciyu Ruan, Haoyang Wang and 6 more

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

– 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.

88%Must read

?Must readVote to see the score

NeurIPS 2026The University of Tokyo Tsinghua University, Tsinghua Tsinghua Alignment & preference optimization

Alignment Dynamics in LLM Fine-Tuning

A unified framework decomposes LLM alignment dynamics into competing rebound and driving forces, explaining reversal and faster re-alignment via rehearsal priming.

Yuhan Huang, Huanran Chen, Yinpeng Dong

Sydney Poster Session 5, Thu, Dec 10, 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.

AI panel: 15 of 20 reviewers recommend it

lenient 5/5

medium 7/10

strict 3/5

71%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Tsinghua Tsinghua University, Tsinghua Graphs & LLMs

A Unified Graph Language Model for Multi-Domain Multi-Task Graph Alignment Instruction Tuning

UniGraphLM proposes a unified graph language model that multi-domain multi-task aligns GNN representations to LLMs via adaptive alignment for cross-domain generalization.

Haibo Chen, Xin Wang, Jiaheng Chao, Ling Feng and 1 more

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

– ReadersNo votes yet

6/20 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 2026Tsinghua Li Auto Inc.JD.com Horizon Robotics Tsinghua University, Tsinghua Deep RL

CRAFT: Counterfactual-to-Interactive Reinforcement Fine-Tuning for Driving Policies

CRAFT combines dense counterfactual advantages with grounded residual corrections to reduce variance and bias in closed-loop autonomous driving fine-tuning, achieving strong Bench2Drive gains.

Keyu Chen, Nanfei Ye, Yida Wang, Wenchao Sun 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.

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Tsinghua Tsinghua-Berkeley Shenzhen Insti Peking Zhejiang University of Technolog Tsinghua University, Tsinghua Multimodal reasoning

CAVE: A Structured Credit Assignment Approach for Fragmented Visual Evidence Reasoning

CAVE uses structured process rewards to improve vision-language models' integration of nonlocal visual evidence, boosting fragmented reasoning benchmarks.

Tengda Guo, Jie Leng, Hanlei Li, Yaoyuan Liang and 5 more

Sydney Poster Session 4, Wed, Dec 9, 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.

74%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Tsinghua University, Tsinghua Hunan Dalian Martime The Chinese University of Hong K Tsinghua Unified multimodal models

DCVD: Dual-Channel Cross-Modal Fusion for Joint Vulnerability Detection and Localization

DCVD jointly detects vulnerable functions and localizes vulnerable statements via dual-channel cross-modal fusion with explicit multi-granularity supervision, outperforming state-of-the-art baselines.

Wenxin Tang, Wenbin Li, Junliang Liu, Jingyu Xiao and 9 more

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

– 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.

91%Must read

?Must readVote to see the score

NeurIPS 2026Tsinghua JD.com Tsinghua University, Tsinghua Beijing Institute of Technology Tianjin Vision-language models

MM-IssueLoc: A Controlled Benchmark for Evaluating Visual Evidence in Multimodal Repository-Level Issue Localization

MM-IssueLoc benchmarks multimodal repository-level issue localization using visual evidence across 652 instances, showing current systems achieve under 39% file accuracy and text-only scores do not transfer.

Shaoxiong Zhan, Shi Hu, Hai Lin, BoyuFeng and 6 more

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

89%Must read

?Must readVote to see the score

NeurIPS 2026Wuhan Harbin Institute of Technology, Tsinghua University, Tsinghua Tsinghua Alignment & preference optimization

Alignment Imprint: Zero-Shot AI-Generated Text Detection via Provable Preference Discrepancy

Deriving an alignment imprint from LLM preference tuning, LAPD detects AI-generated text with 45.82% relative gains over baselines via statistically guaranteed preference discrepancy.

Junxi Wu, Kailin Huang, Dongjian Hu, Bin Chen and 3 more

Sydney Poster Session 1, Tue, Dec 8, 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.

89%Must read

?Must readVote to see the score

NeurIPS 2026SUN YAT-SEN UNIVERSITY Chinese University of Hong Kong The Chinese University of Hong K Nanjing University of Aeronautic Southwest Jiaotong Multimodal reasoning

AgroOmni: A Large-Scale Multi-view Agricultural Dataset for Cross-Scale Multimodal Reasoning

AgroOmni introduces 288K multi-view agricultural VQA pairs across scales, and AgroNVILA achieves state-of-the-art 62.32% on AgroMind while demonstrating strong cross-scale generalization.

Jiarui Zhang, Junqi Hu, Zurong Mai, Yang Liu and 9 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.

80%Must read

?Must readVote to see the score

NeurIPS 2026U Toronto Tsinghua The Chinese University of Hong K Tsinghua University, Tsinghua GigaAI Embodied agents

DriveDreamer-Policy: A Geometry-Grounded World-Action Model for Unified Generation and Planning

DriveDreamer-Policy unifies depth generation, video prediction, and motion planning via geometry-aware world representations, achieving 89.2 PDMS on Navsim v1 and 88.7 EPDMS on v2.

Yang Zhou, Xiaofeng Wang, Hao Shao, Letian Wang and 7 more

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 2 on Hugging Face · Code ★ 60

– 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 2026McGill Northeastern Imperial College London King's College London, Univ Tsinghua Robot manipulation

MyoChallenge 2025: A New Benchmark for Human Athletic Intelligence

MyoChallenge 2025 benchmarks musculoskeletal sports control via simulated table tennis and soccer tasks, advancing agile motor algorithms across 70 teams.

Cheryl Wang, Chun Kwang Tan, Balint Hodossy, Shirui Lyu and 20 more

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 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.

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Tsinghua University, Tsinghua Department of Industrial Enginee Tsinghua Huawei Technologies Ltd.Datadog Anomaly detection

VETime: Vision Enhanced Zero-Shot Time Series Anomaly Detection

VETime unifies temporal and visual modalities via fine-grained alignment and dynamic fusion for zero-shot time-series anomaly detection, outperforming state-of-the-art models with lower overhead.

Yingyuan Yang, Tian Lan, Yifei Gao, Yimeng Lu and 5 more

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 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.

72%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Tsinghua Department of Automation, Tsingh Tsinghua University, Tsinghua U Maryland, College Park Independent Researcher RL for LLMs

Listwise Policy Optimization: Group-based RLVR as Target-Projection on the LLM Response Simplex

Listwise Policy Optimization explicitly projects policies onto target distributions over response simplices via divergence minimization, improving reasoning performance and stability over group-based policy gradients.

Yun Qu, Qi Wang, Yixiu Mao, Heming Zou and 10 more

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 67 on Hugging Face

– 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.

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Independent Gradient Network U Science and Technology of Chin Tsinghua University, Tsinghua U Edinburgh, University of Edinb RL for LLMs

ECHO-2: A Large-Scale Distributed Rollout Framework for Cost-Efficient Reinforcement Learning

ECHO-2 is a distributed RL framework that overlaps rollout generation, dissemination, and training with bounded policy staleness to improve cost efficiency while preserving rewards.

Jingwei Song, Meng Chen, Jie Xiao, Qingnan Ren and 14 more

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 13 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.

91%Must read

?Must readVote to see the score

NeurIPS 2026The Hong Kong University of Scie Alibaba Group Renmin University of China The Chinese University of Hong K Peking LLM agents & planning

Towards On-Policy Data Evolution for Visual-Native Multimodal Deep Search Agents

A visual-native harness with an image bank and on-policy data evolution improves multimodal deep search agents, raising Qwen3-VL-8B to 39.0% average and surpassing Gemini-2.5 Pro.

Shijue Huang, Hangyu Guo, Guanting Dong, Chenxin Li and 7 more

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 21 on Hugging Face · Code ★ 30

– 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.

89%Must read

?Must readVote to see the score

NeurIPS 2026Tencent Harbin Institute of Technology Hong Kong Polytechnic Westlake Jiangnan Alignment & preference optimization

Aligning LLMs with Biomedical Knowledge using Balanced Fine-Tuning

Balanced Fine-Tuning uses dual-scale token and sequence reweighting targeting dense epistemic uncertainty to align LLMs with biomedical knowledge, improving reasoning and sparse-reward RL over standard fine-tuning.

Zhenchao Tang, Fang Wang, Haohuai He, Jiale Zhou and 12 more

Sydney Poster Session 5, Thu, Dec 10, 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.

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Tsinghua University, Tsinghua Tsinghua Tsinghua University, Microsoft Xiaomi Corporation Microsoft Research Asia LLM pretraining & scaling laws

Memory Grafting: Scaling Language Model Pre-training via Offline Conditional Memory

Memory Grafting uses frozen hidden states from a grafting model as conditional n-gram memory for language models, improving average benchmarks to 53.86 versus 52.43 for vanilla Engram at 2.8B scale with minimal overhead.

Runxi Cheng, Yuchen Guan, Yongxian Wei, Qianpu Sun and 6 more

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 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.

83%Must read

?Must readVote to see the score

NeurIPS 2026Tsinghua Tsinghua University, Tsinghua StableAI Southeast Siemens Corporate Research Diffusion models

Breaking the Quality–Privacy Tradeoff in Tabular Data Generation via In-Context Learning

DiffICL frames tabular synthesis as in-context learning using pretrained structural priors to avoid memorization, improving both quality and privacy in small-data settings.

Xinyan Han, yan Lu, Xiaoyu Lin, Yuanyuan Jiang and 4 more

Sydney Poster Session 6, Thu, Dec 10, 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 2026Chinese Academy of Sciences Tsinghua Tsinghua University, Tsinghua 3D vision & reconstruction

CORAL: A Benchmark for Structure-aware and Brain-wide Neuron Reconstruction in Light Microscopy

CORAL introduces a structure-aware benchmark for local and whole-brain neuron reconstruction in light microscopy, revealing current methods need improvement for complete brain-wide tracing.

Zekang Yang, Jiamin Li, Zhenghua Li, Jiaqi Fan and 2 more

Sydney Poster Session 3, Wed, Dec 9, 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.

83%Must read

?Must readVote to see the score

NeurIPS 2026Tsinghua Tsinghua University, Tsinghua Meta Vision-language-action models

Generative Control as Optimization: Time Unconditional Flow Matching for Adaptive and Robust Robotic Control

GeCO replaces fixed-schedule flow matching with time-unconditional optimization that adaptively allocates inference compute and uses field norms as training-free OOD detectors.

Zunzhe Zhang, Runhan Huang, Yicheng Liu, Shaoting Zhu and 2 more

Atlanta Poster Session 5, Fri, Dec 11, 10:00 AM–1:00 PM, Hall C1 · 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.

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Tsinghua Tsinghua University, Tsinghua Google , Tsinghua Biosignals & wearables

FacePhys: State of the Heart Learning

FacePhys is a memory-efficient rPPG algorithm using temporal-spatial state space duality that cuts error by 49% with a 3.6 MB footprint and 9.46 ms latency.

Jack Tang, Kegang Wang, Yuntao Wang, Xin Liu and 4 more

Paris Poster Session 4, Thu, Dec 10, 5:30 PM–7:30 PM, Paris Poster Hall · 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 6/10

strict 1/5

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Beihang Beijing University of Aeronautic School of Computer Science and E Apple Bytedance Multi-agent RL

Heterogeneous Agent Collaborative Reinforcement Learning

HACRL enables heterogeneous agents to share verified rollouts during collaborative on-policy training and execute independently at inference, with HACPO improving all agents by 3.6% over baselines at half the rollout cost.

Zhixia Zhang, Zixuan Huang, Gonxun Li, Huaiyang Wang and 8 more

Paris Poster Session 5, Fri, Dec 11, 11:30 AM–1:30 PM, Paris Poster Hall · Published 2026 · ▲ 110 on Hugging Face

– 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 2026U Illinois Urbana-Champaign Tsinghua University, Tsinghua U Illinois at Urbana-Champaign Citadel Securities Department of Computer Science, Agent memory

Useful Memories Become Faulty When Continuously Updated by LLMs

LLM-updated agent memories degrade with repeated consolidation, often dropping below baselines; retaining raw episodic traces doubles accuracy versus forced consolidation.

Dylan Zhang, Yanshan Lin, Zhengkun Wu, Yihang Sun and 3 more

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026 · ▲ 18 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.

89%Must read

?Must readVote to see the score

NeurIPS 2026Tsinghua University, Tsinghua Tsinghua Xidian U Illinois Urbana-Champaign YuanXingGuangNian Vision-language-action models

ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models

ElegantVLA accelerates vision-language-action models via adaptive compute scheduling, achieving up to 3.77x speedup and doubling control frequency without retraining.

Ye Li, Huanan Liu, Kangye Ji, Yuan Meng and 6 more

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 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 10/10

strict 1/5

86%Must read

?Must readVote to see the score

NeurIPS 2026Technical University of Munich Tsinghua University, Tsinghua Alibaba Group Ant Group Ant Financial LLM evaluation & benchmarks

DiffScore: Text Evaluation Beyond Autoregressive Likelihood

DiffScore evaluates text with masked diffusion models using bidirectional context to eliminate positional bias and decompose quality into fluency and faithfulness, outperforming autoregressive baselines.

Wen Lai, Yingli Shen, Dingnan Jin, Qing Cui and 3 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.

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026OralU Cambridge Southern University of Science a Tsinghua University, Tsinghua IMATI-CNR Nanjing 3D vision & reconstruction

SAM 3D Animal: Promptable Animal 3D Reconstruction from Images in the Wild

SAM 3D Animal is a promptable framework using the SMAL+ model and Herd3D dataset to reconstruct multiple 3D animals from single images with keypoint and mask prompts, achieving state-of-the-art results.

Xuyi Hu, Jin Lyu, Jiuming Liu, Yebin Liu and 3 more

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 3 on Hugging Face · Code ★ 13

– 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 6/10

strict 1/5

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026ByteDance Inc.SUN YAT-SEN UNIVERSITY Tsinghua Tsinghua University, Tsinghua Applied Research Center, Tencent Video generation

Temporal Concentration from Rollout Errors: Implicit Preference Optimization For Text-to-Video Diffusion

cIPO aligns text-to-video diffusion by deriving implicit preferences from reconstruction errors and concentrating optimization on high-error temporal segments to fix sparse artifacts.

henglin liu, Fangyuan Kong, Jing Wang, Yizhou Lin and 6 more

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 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.

AI panel: 10 of 20 reviewers recommend it

lenient 4/5

medium 6/10

strict 0/5

80%Must read

?Must readVote to see the score

NeurIPS 2026China Telecom China Telecom Artificial Intelli Tianjin Tsinghua University, Tsinghua Spoken language models

TELEVAL: A Benchmark Designed for Spoken Language Models in Chinese Interactive Scenarios

TELEVAL benchmarks Chinese spoken language models on interactional audio-conditioned dialogue, finding competitive semantic accuracy but degraded interactional performance under acoustic variability and a recurring caption trap failure.

Zehan Li, Hongjie Chen, Qing Wang, Yuxin 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

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

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026OralShanghai Jiao Tong Huazhong University of Science a Tsinghua University, Tsinghua U California, San Diego Carnegie Mellon Unified multimodal models

LatentUM: Unleashing the Potential of Interleaved Cross-Modal Reasoning via a Latent-Space Unified Model

LatentUM unifies modalities in a shared latent space to enable efficient interleaved cross-modal reasoning and generation, achieving state-of-the-art visual planning and self-reflective generation results.

Jiachun Jin, Zetong Zhou, Xiao Yang, Hao Zhang and 3 more

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 29 on Hugging Face · Code ★ 63

– 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

83%Must read

?Must readVote to see the score

NeurIPS 2026Tsinghua Beijing University of Post and T Tsinghua University, Tsinghua Vision-language models

SwiftVLM: Efficient Vision-Language Model Inference via Cross-Layer Token Bypass

SwiftVLM introduces cross-layer token bypass to preserve visual tokens across pruning stages, enabling training-free vision-language model acceleration with superior accuracy-efficiency trade-offs.

Qian, Xinran Yu, Xinan Wang, Danyang Li and 4 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.

AI panel: 13 of 20 reviewers recommend it

lenient 4/5

medium 8/10

strict 1/5

80%Must read

?Must readVote to see the score

NeurIPS 2026tsinghua university JD.com Tsinghua University, Tsinghua School of Software, Tsinghua BNRist, Tsinghua Diffusion models

RA-CFGCache: From Branch-Level Criteria to Guided-Risk Control under Classifier-Free Guidance

RA-CFGCache aligns caching risk with CFG branch errors and timestep propagation to improve diffusion model efficiency and fidelity.

Yiming Liu, Ben Wan, Hui Chen, Ao Wang and 4 more

Sydney Poster Session 1, Tue, Dec 8, 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 3/5

medium 8/10

strict 1/5

88%Must read

?Must readVote to see the score

NeurIPS 2026National University of Singapore Tsinghua National university of Singapore Suzhou Automobile Research Insti Meta Agent benchmarks & environments

FML-bench: A Controlled Study of AI Research Agent Strategies from the Perspective of Search Dynamics

FML-bench isolates agent strategy from infrastructure across 18 ML tasks, finding greedy hill-climbing nearly matches tree search, while adaptive exploration switching outperforms fixed strategies.

Qiran Zou, Hou Hei Lam, Wenhao Zhao, Tingting Chen and 10 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.

AI panel: 15 of 20 reviewers recommend it

lenient 5/5

medium 9/10

strict 1/5

91%Must read

?Must readVote to see the score

NeurIPS 2026Tsinghua University, Tsinghua Xiaobing.AI Peking Advanced Micro Devices Tsinghua University, Microsoft Embodied agents

TaskGround: Structured Executable Task Inference for Full-Scene Household Reasoning

TaskGround grounds full household scenes into task-relevant slices to infer executable task structures, improving compact open-weight models' success rates by large margins over direct prompting while cutting token costs up to 18x.

ZhiYuan Feng, Yu Deng, Ruichuan An, Zhenhua Liu and 10 more

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

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026The Hong Kong University of Scie Tsinghua University, Tsinghua RoboScience Technical University of Munich ByteDance Inc.Image restoration

IRPO: Boosting Image Restoration via Post-training GRPO

IRPO applies GRPO post-training to image restoration via selective hard-sample data and multi-component rewards, improving in-domain accuracy by 0.93 dB and OOD generalization by 3.43 dB.

Haoxuan Xu, Yi Liu, Tianfu Li, Ruolin Shen and 6 more

Paris Poster Session 2, Wed, Dec 9, 5:00 PM–7:00 PM, Paris Poster Hall · 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.

AI panel: 10 of 20 reviewers recommend it

lenient 4/5

medium 5/10

strict 1/5

86%Must read

?Must readVote to see the score

NeurIPS 2026Tsinghua Institute for AI Industry Resear Tsinghua University, Tsinghua Genomics & single-cell

CellMSA: Context Modeling for Single-Cell Representation Learning

CellMSA improves single-cell representation learning by modeling cross-batch and cross-cell-type context via MSA-inspired gene-pair representations, outperforming existing methods across benchmarks.

Suyuan Zhao, Minghao Liu, Yizhen Luo, Zaiqing Nie

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 5/5

medium 8/10

strict 1/5

80%Must read

?Must readVote to see the score

NeurIPS 2026Tsinghua The Chinese University of Hong K Ant Group Tsinghua University, Tsinghua Renmin University of China LLM pretraining & scaling laws

CausalMix: Data Mixture as Causal Inference for Language Model Training

CausalMix frames data mixture optimization as causal inference to dynamically estimate optimal mixtures via conditional average treatment effects, improving LLM performance without retraining proxy models.

Zinan Tang, YUKUN ZHANG, Shaomian Zheng, Zhuoshi Pan and 5 more

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 19 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.

AI panel: 12 of 20 reviewers recommend it

lenient 5/5

medium 7/10

strict 0/5

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Tsinghua Carnegie Mellon University / IST U Roma "La Sapienza"Carnegie Mellon Tsinghua University, Tsinghua Long-context modeling

DashAttention: Differentiable and Adaptive Sparse Hierarchical Attention

DashAttention uses adaptive α-entmax to select variable key-value blocks per query, enabling fully differentiable hierarchical sparse attention that matches full-attention accuracy at 75% sparsity with faster inference than FlashAttention-3.

Yuxiang Huang, Nuno Gonçalves, Federico Alvetreti, Lei Li and 4 more

Paris Poster Session 6, Fri, Dec 11, 2:30 PM–4:30 PM, Paris Poster Hall · 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.

AI panel: 10 of 20 reviewers recommend it

lenient 4/5

medium 6/10

strict 0/5

74%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026National University of Singapore Tsinghua University, Tsinghua U Science and Technology of Chin Tsinghua Multimodal reasoning

UniVLR: Unifying Text and Vision in Visual Latent Reasoning for Multimodal LLMs

UniVLR unifies text and visual reasoning into a shared visual workspace, using compressed visual latent tokens to outperform prior methods with fewer reasoning tokens.

Houcheng Jiang, Jiajun Fu, Junfeng Fang, Chen Gao and 3 more

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

– 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

83%Must read

?Must readVote to see the score

NeurIPS 2026Hong Kong University of Science The Hong Kong University of Scie Singapore University of Technolo Nanyang Technological SophiaPro LLM agents & planning

Harnessing Agentic Evolution

AEvo formulates agentic evolution as an interactive environment where a meta-agent edits the evolution procedure to steer long-horizon search, achieving up to 26% relative improvement over baselines.

Jiayi Zhang, Yongfeng Gu, Jianhao Ruan, Maojia Song and 8 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.

AI panel: 13 of 20 reviewers recommend it

lenient 4/5

medium 7/10

strict 2/5

80%Must read

?Must readVote to see the score

NeurIPS 2026Tsinghua Sapient Intelligence Tsinghua University, Tsinghua Ragentile Intelligence Deep RL

Signal-Adaptive Trust Regions for Gradient-Free Optimization of Recurrent Spiking Neural Networks

SATR constrains gradient-free RSNN updates via signal-adaptive KL trust regions, improving stability and matching PPO-LSTM returns on continuous control benchmarks.

Jinhao Li, Yuhao Sun, Zhiyuan Ma, Hao He and 4 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

80%Must read

?Must readVote to see the score

NeurIPS 2026Nanyang Technological Department of Computer Science, U Illinois Chicago Tsinghua University, Tsinghua Agent memory

MemSkill: Learning and Evolving Memory Skills for Self-Evolving Agents

MemSkill learns and evolves reusable memory skills for extracting and revising agent memories via selection, execution, and design loops, improving long-context task performance.

Haozhen Zhang, Quanyu Long, Jianzhu Bao, Tao Feng and 3 more

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 64 on Hugging Face · Code ★ 582

– 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

86%Must read

?Must readVote to see the score

NeurIPS 2026Institute for AI Industry Resear Tsinghua Institute for AI Industry Resear Tsinghua University, Tsinghua Proteins

Structure-aware Reinforcement Learning for Protein Directed Evolution

StructEvo uses structure-aware reinforcement learning with delta-structure fusion and hierarchical actions to outperform state-of-the-art protein directed evolution methods by up to 16.3%.

Zikun Nie, Suyuan Zhao, Yizhen Luo, Siqi Fan and 1 more

Sydney Poster Session 3, Wed, Dec 9, 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

83%Must read

?Must readVote to see the score

NeurIPS 2026Tsinghua Univerisity Tsinghua Z.AI Co., Ltd.Tsinghua University, Tsinghua Z.AI Web & GUI agents

SCALECUA : Scaling Computer Use Agents with Verifiable Task Synthesis and Efficient Online RL

ScaleCUA scales computer use agents via verifiable synthetic tasks and efficient online RL, achieving new open-source state-of-the-art results on OSWorld and ScienceBoard.

Bowen Lv, Xiao Liu, Yanyu Ren, Hanyu Lai and 6 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.

AI panel: 13 of 20 reviewers recommend it

lenient 5/5

medium 7/10

strict 1/5

91%Must read

?Must readVote to see the score

NeurIPS 2026OralU California, Berkeley Tsinghua Tsinghua University, Tsinghua Purdue U Washington Agent benchmarks & environments

MLS-Bench: A Holistic and Rigorous Assessment of AI Systems on Building Better AI

MLS-Bench evaluates AI agents on inventing scalable ML methods across 140 tasks, finding current systems fail to reliably surpass human-designed approaches due to insufficient scientific validation insight.

Bohan Lyu, Yucheng Yang, Siqiao Huang, Jiaru Zhang and 24 more

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026 · ▲ 8 on Hugging Face · Code ★ 120

– 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 8/10

strict 4/5

86%Must read

?Must readVote to see the score

NeurIPS 2026Tsinghua Shandong Beijing Institute of Technology Tsinghua University, Tsinghua Alibaba Group Embodied agents

WorldVLN: Autoregressive World Action Model for Aerial Vision-Language Navigation

WorldVLN autoregressively predicts short-horizon world-state transitions to generate waypoint actions for aerial vision-language navigation, achieving over 12% success-rate gains and real-world drone transfer.

Baining Zhao, jiacheng xu, Weicheng Feng, Xin Zhang and 12 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

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

88%Must read

?Must readVote to see the score

NeurIPS 2026Carnegie Mellon Shanghai Jiaotong Tsinghua University, Tsinghua MIT, NVIDIA NVIDIA Long-context modeling

Jet-Long: Efficient Long-Context Extension with Dynamic Bifocal RoPE

Jet-Long dynamically rescales RoPE via bifocal local and long-range windows with an analytic length-aware schedule to extend LLM contexts without tuning, outperforming baselines on RULER, HELMET-RAG, and perplexity while retaining near-FlashAttention-3 throughput.

Haozhan Tang, Zerui Wang, Yuxian Gu, Song Han and 1 more

Atlanta Poster Session 4, Thu, Dec 10, 4:30 PM–7:30 PM, Hall C1 · Published 2026 · ▲ 24 on Hugging Face · Code ★ 13

– 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