Good Papers: read the good ones

PAGER closes the semantic-execution gap for point-precise geometric GUI control via dependency-structured planning and pixel-level execution, achieving 4.1x higher task success than general baselines.

Jingxuan Wei, Xi Bai, Shan Liu, caijun jia and 7 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 15 on Hugging Face · Code ★ 3

– 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

80%Must read

?Must readVote to see the score

NeurIPS 2026U the Chinese Academy of Science Institute of Software, Chinese A Beijing University of Aeronautic Institute of Software, CAS U Chinese Academy of Sciences RL for LLMs

Verifiable Environments Are LEGO Bricks: Recursive Composition for Reasoning Generalization

RACES recursively composes verifiable environments as LEGO bricks to scale RL reasoning training, boosting model performance on unseen benchmarks with far fewer base environments.

Hao Xiang, Qiaoyu Tang, Le Yu, Yaojie Lu and 7 more

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

medium 8/10

strict 0/5

80%Must read

?Must readVote to see the score

NeurIPS 2026Waseda U the Chinese Academy of Science Alignment & preference optimization

IRIS: Interpolative Rényi Iterative Self-play for Large Language Model Fine-Tuning

IRIS unifies self-play fine-tuning via adjustable Rényi divergence with adaptive schedules, surpassing supervised fine-tuning with fewer annotations across benchmarks.

Wenjie Liao, Like Wu, Liangjie Zhao, Shihui Xu and 1 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.

72%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U the Chinese Academy of Science U Chinese Academy of Sciences Knowledge editing

MetaKE: Meta-Learning for Knowledge Editing Toward a Better Accuracy-Editability Trade-off

MetaKE unifies knowledge editing stages via bi-level optimization with structural gradient proxies, improving the accuracy-editability trade-off.

Shuxin Liu, Di Gao, Ou Wu

Sydney Poster Session 5, Thu, Dec 10, 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.

AI panel: 8 of 20 reviewers recommend it

lenient 3/5

medium 5/10

strict 0/5

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Institute of Software, Chinese A U the Chinese Academy of Science Institute of Software Chinese Ac Vision-language-action models

PAPO-VLA: Planning-Aware Policy Optimization for Vision-Language-Action Models

PAPO-VLA improves vision-language-action reliability by identifying planning actions via action variation and trajectory outcomes, weighting them by causal importance in policy optimization, and boosting benchmark performance.

Peizheng Guo, Jingyao Wang, Changwen Zheng, Wenwen Qiang

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.

AI panel: 11 of 20 reviewers recommend it

lenient 5/5

medium 6/10

strict 0/5

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Huawei Technologies Ltd.City University of Hong Kong Noah Ark Lab U the Chinese Academy of Science Huawei Noah's Ark Lab Test-time compute

PSD: Pushing the Pareto Frontier of Diffusion LLMs via Parallel Speculative Decoding

PSD accelerates diffusion LLM inference via adaptive parallel unmasking and multi-depth speculative drafts with hierarchical verification, achieving up to 5.5x tokens per pass with near-greedy accuracy.

Shengyin Sun, Yiming Li, Renxi Liu, Xinqi Li and 6 more

Sydney Poster Session 2, Tue, Dec 8, 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.

AI panel: 10 of 20 reviewers recommend it

lenient 4/5

medium 4/10

strict 2/5

74%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026nanjing university Kuaishou- 快手科技 Nanjing university U the Chinese Academy of Science Nanjing Video-language models

ViDiC: Video Difference Captioning

ViDiC introduces a video difference captioning task and ViDiC-1K benchmark that reveals large multimodal models struggle with fine-grained comparative video perception.

Jiangtao Wu, Shihao Li, Zhaozhou Bian, Jialu Chen and 6 more

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 29 on Hugging Face · Code ★ 15

– 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.

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

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Alibaba Alibaba Group U the Chinese Academy of Science Alibaba Inc.Controllable generation

FashionChameleon: Towards Real-Time and Interactive Human-Garment Video Customization

FashionChameleon enables real-time interactive multi-garment video customization via teacher in-context learning, streaming distillation, and KV cache rescheduling, achieving 23.8 FPS and 30-180x speedups.

Quanjian Song, Yefeng Shen, Mengting Chen, Hao Sun and 3 more

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 65 on Hugging Face · Code ★ 284

– 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

71%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Institute of Computing Technolog Northwestern Research Center Of Distributed S U the Chinese Academy of Science Institute of Computing Technolog Graph generation

GraDE: A Graph Diffusion Estimator for Frequent Subgraph Discovery in Neural Architectures

GraDE uses a graph diffusion estimator to score subgraph typicality and discovers large-scale neural architecture motifs with up to 30x higher median frequency than sampling methods.

Yikang Yang, Zhengxin Yang, Minghao Luo, Luzhou Peng and 4 more

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 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.

76%Highly rated

NeurIPS 2026SpotlightInstitute of Information Enginer Institute of Information Enginee JD.com U the Chinese Academy of Science Institute of Information Enginee RL for LLMs

Self-Distilled RLVR

RLSD combines RLVR and self-distillation, using token-level policy differences for update magnitudes and environmental feedback for directions, improving convergence and stability.

Chenxu Yang, Chuanyu Qin, Qingyi Si, Minghui Chen and 6 more

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 145 on Hugging Face

– ReadersNo votes yet

10/20 AI panelreviewers recommend it

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

AI panel: 10 of 20 reviewers recommend it

lenient 3/5

medium 6/10

strict 1/5

80%Must read

?Must readVote to see the score

NeurIPS 2026East China Normal JD.com U the Chinese Academy of Science Institute of Information Enginer Institute of Information Enginee Video-language models

Harnessing Streaming Video in the Wild

Streaming-Train-248K and Streaming Harness adapt VLMs to real-time video streams with proactive interaction, 12-hour memory, and sub-second latency.

Dingyu Yao, Shuhuan Gu, Qingyi Si, Junhao Zhou and 7 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 5/5

medium 5/10

strict 2/5

80%Must read

?Must readVote to see the score

NeurIPS 2026U Chinese Academy of Sciences Johns Hopkins Peking U the Chinese Academy of Science Zhejiang Vision-language models

dRAE: Representation Autoencoder with Hyper-Spherical Codes

Hyper-spherical quantization decouples semantics from magnitude via angular routing to prevent codebook collapse, enabling scalable discrete representation autoencoders with full codebook usage and high-fidelity reconstruction.

Tianren Ma, Lin Long, Chuyan Chen, Mu Zhang and 3 more

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

– 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.

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U the Chinese Academy of Science Tianjin U Chinese Academy of Sciences Instruction tuning

One Algorithm, Two Goals: Dual Scoring for Parameter and Data Selection in LLM Fine-Tuning

A shared gradient-interaction scoring rule unifies parameter and data selection for LLM fine-tuning via DualSFT, improving joint efficiency and trade-offs.

Xinrui Chen, Liu Yang, Ou Wu

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

83%Must read

?Must readVote to see the score

NeurIPS 2026JD.com U the Chinese Academy of Science JD.COM Nankai U the Chinese Academy of Science Instruction tuning

The Hidden Power of Scaling Factor in LoRA Optimization

LoRA's scaling factor dominates optimization by amplifying task signals without increasing drift, outperforming learning rate adjustments. The optimal alpha follows a sublinear square-root law with rank, revealing insufficient scaling in existing heuristics. Proposed LoRA-alpha restores principled s

Zicheng Zhang, Haoran Li, Jiaxing Wang, Guoqiang Gong and 9 more

Paris Poster Session 5, Fri, Dec 11, 11:30 AM–1:30 PM, Paris Poster Hall · Published 2026 · ▲ 12 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 3/5

medium 8/10

strict 2/5

80%Must read

?Must readVote to see the score

NeurIPS 2026U the Chinese Academy of Science Tsinghua Tencent HY Institute of Information Enginee Nankai Visual grounding

Learn where to Click from Yourself: On-Policy Self-Distillation for GUI Grounding

GUI-SD uses on-policy self-distillation with privileged visual contexts and entropy-guided distillation for GUI grounding, outperforming GRPO methods in accuracy and efficiency.

Yan Zhang, Daiqing Wu, Huawen Shen, Can Ma and 1 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 5 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

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U the Chinese Academy of Science Nankai Guizhou Power Grid Co., Ltd.Institute of Software, Chinese A Institute of Software Chinese Ac Time series forecasting

Dirichlet-Guided Group Forecasting for Alleviating Over-smoothing in Time Series Forecasting

Dirichlet-Guided Group Forecasting reduces time-series over-smoothing by modeling multi-modal predictive distributions with Dirichlet-guided sampling, improving accuracy, diversity, and dynamical consistency.

Xingyu Zhang, Jingyao Wang, Xin Yu, Zeen Song and 3 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.

AI panel: 11 of 20 reviewers recommend it

lenient 4/5

medium 7/10

strict 0/5

72%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Sichuan U the Chinese Academy of Science Peking , Chinese Academy of Sciences Institute for AI Industry Resear Materials & chemistry

AtomWorld-Mem: Memory-Restored World States for Long-Horizon Atomistic Evolution

AtomWorld-Mem restores latent hidden dynamical states from atomistic snapshots via multi-scale memory to improve long-horizon kinetic Monte Carlo evolution and transfer across unseen alloys.

Tian Luo, Ruge Zhang, Haozhi Han, Yifeng Chen and 4 more

Sydney Poster Session 6, Thu, Dec 10, 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.

AI panel: 8 of 20 reviewers recommend it

lenient 3/5

medium 5/10

strict 0/5

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U the Chinese Academy of Science Institute of Computing Technolog Chinese Academy of Sciences Institute of Computing Technolog Meta, transfer & few-shot learning

Dynamical Adapter Fusion: Constructing A Global Adapter for Pre-Trained Model-based Class-Incremental Learning

Dynamical Adapter Fusion derives optimal coefficients via PAC-Bayes and Taylor expansion to fuse task-specific adapters into one global adapter, achieving state-of-the-art class-incremental learning results.

RuiQi Liu, Boyu Diao, Zijia An, Runjie Shao and 4 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.

AI panel: 11 of 20 reviewers recommend it

lenient 4/5

medium 7/10

strict 0/5

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Academy of Mathematics and Syste U the Chinese Academy of Science Ohio State University, Columbus JD.com CAIR, HKISI, CAS Optimization

Adam under Generalized Smoothness with Second-Moment-Type Stochastic Gradients

Adam converges with high probability on generalized-smooth objectives under only second-moment stochastic gradients, matching a sharp δ^{-1/2} confidence dependence and yielding expectation rates for p<1.

ruinan Jin, Difei Cheng, Ling Chen, Jun Luo and 2 more

Sydney Poster Session 6, Thu, Dec 10, 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 2/5

medium 6/10

strict 3/5

83%Must read

?Must readVote to see the score

NeurIPS 2026Institute of Information Enginee U the Chinese Academy of Science Institute of Information Enginee Controllable generation

NP-LoRA: Null Space Projection for Subject-Style LoRA Fusion

NP-LoRA fuses subject and style LoRAs via null-space projection to reduce subspace interference, improving controllable generation without retraining.

Chuheng Chen, Xiaofei Zhou, Geyuan Zhang, Yong Huang and 1 more

Sydney Poster Session 5, Thu, Dec 10, 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 5/5

medium 7/10

strict 1/5

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Institute of Information Enginee Institute of Information Enginee U the Chinese Academy of Science Jiangxi University of Finance an SAIT-China Lab, Samsung Research LLM reasoning & chain of thought

Mitigating Overthinking in Large Reasoning Language Models via Reasoning Path Deviation Monitoring

A reasoning path deviation metric detects high-entropy overthinking tokens to dynamically terminate redundant reasoning, improving performance and efficiency over existing early-exit methods.

Weixin Guan, Liang Li, Jiapeng Liu, Bing Li and 5 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.

AI panel: 11 of 20 reviewers recommend it

lenient 4/5

medium 7/10

strict 0/5

74%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U Science and Technology of Chin U the Chinese Academy of Science City University of Hong Kong Tsinghua Zhejiang Embodied agents

MineEvolve: Self-Evolution with Accumulated Knowledge for Long-Horizon Embodied Minecraft Agents

MineEvolve converts Minecraft execution feedback into structured skills and remedies via Monitor, Inducer, Curator, and Adaptor, improving long-horizon agent performance across planners.

Zhengwei Xie, Zhisheng Chen, Ziyan Weng, Jinhan Li and 6 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.

AI panel: 9 of 20 reviewers recommend it

lenient 4/5

medium 5/10

strict 0/5