Good Papers: read the good ones

A neural framework learns pole-residue modal decomposition from system observables alone, generalizing to unseen port counts and recovering physical eigenmodes without modal supervision.

Juho Park, Kaushik Sengupta

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

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026UC Berkeley Princeton U Toronto & Vector Institute Flatiron Institute, Simons Found Optimization

Sharp Capacity Scaling of Spectral Optimizers in Learning Associative Memory

Spectral optimizer Muon exceeds SGD associative memory capacity, matching Newton's method with first-order updates and larger critical batch sizes.

Juno Kim, Eshaan Nichani, Denny Wu, Alberto Bietti and 1 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.

AI panel: 10 of 20 reviewers recommend it

lenient 3/5

medium 4/10

strict 3/5

91%Must read

?Must readVote to see the score

NeurIPS 2026UCSB Princeton Google DeepMind Mila, University of Montreal U Oxford Vision-language models

TokenSwap: Benchmarking and Reducing the Modality Gap in Multimodal LLMs

TokenSwap benchmarks and reduces MLLMs' modality gap by interleaving visual tokens with text, finding reasoning models have smaller gaps and training with TokenSwap mitigates it.

Andong Hua, Colton Bishop, Igor Mordatch, Arian Hosseini and 4 more

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

– ReadersNo votes yet

18/20 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: 18 of 20 reviewers recommend it

lenient 4/5

medium 9/10

strict 5/5

72%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Princeton Princeton Multi-agent RL

Self-Supervised Goal-Reaching Results in Multi-Agent Cooperation and Exploration

Self-supervised goal-reaching enables multi-agent cooperation and exploration via sparse feedback, outperforming alternatives and discovering nontrivial coordination without explicit mechanisms.

Chirayu Nimonkar, Shlok Shah, Catherine Ji, Benjamin Eysenbach

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 4/5

medium 3/10

strict 1/5

80%Must read

?Must readVote to see the score

NeurIPS 2026School of Engineering and Applie Harvard MIT Fudan Massachusetts Institute of Techn Agent benchmarks & environments

AI GAMESTORE: Scalable, Open-Ended Evaluation of Machine General Intelligence with Human Games

AI GameStore proposes evaluating general intelligence via scalable synthesis of human games, finding frontier vision-language models score under 10% of human averages on most generated games.

Lance Ying, Ryan Truong, Prafull Sharma, Kaiya Zhao and 8 more

Atlanta Poster Session 6, Fri, Dec 11, 4:30 PM–7:30 PM, Hall C1 · Published 2026 · ▲ 9 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 2025Princeton UT Austin Seoul National MAUM.AI, Soongsil Sirindhorn International Institu Agent benchmarks & environments

The PokeAgent Challenge: Competitive and Long-Context Learning at Scale

PokeAgent is a large-scale Pokémon benchmark with battling and speedrunning tracks that expose major gaps between LLMs, RL agents, and human experts.

Seth Karten, Jake Grigsby, Tersoo Upaa, Junik Bae and 27 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 12 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.

AI panel: 10 of 20 reviewers recommend it

lenient 4/5

medium 6/10

strict 0/5

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026MIT EPFL Princeton EPFL - EPF Lausanne Deep learning theory

Provable Quantization with Randomized Hadamard Transform

Dithered randomized Hadamard quantization is unbiased and achieves mean squared error asymptotically matching dense random rotations at O(d log d) cost.

Ying Feng, Piotr Indyk, Michael Kapralov, Dmitrii Krachun and 1 more

Atlanta Poster Session 6, Fri, Dec 11, 4:30 PM–7:30 PM, Hall C1 · 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 2026Princeton Vision-language models

CoMet: Context and Multiplicity Decomposition for Multimodal Uncertainty Estimation

CoMet decomposes multimodal LLM uncertainty into context and multiplicity terms via a lightweight module, improving calibration without generation or sampling.

Sanghyuk Chun, William Yang, Amaya Dharmasiri, Olga Russakovsky

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

71%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Princeton Rutgers Rutgers - New Brunswick Bayesian & probabilistic methods

Attention as In-Context Empirical Bayes: A Two-Stage View via Particle Dynamics

Attention-only transformers under token corruption implement two-stage in-context empirical Bayes via depth-refined particle dynamics and skip-connection queries, enabling depth-dependent denoising without explicit noise schedules and posterior-mean convergence to Bayes-optimal predictors.

Matthew Smart, Soumya Ganguly, Nilava Metya, Alexandre V Morozov and 1 more

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

AI panel: 6 of 20 reviewers recommend it

lenient 2/5

medium 3/10

strict 1/5

80%Must read

?Must readVote to see the score

NeurIPS 2026Princeton Department of Computer Science, U Toronto NVIDIA Video generation

Addressable Memory for Video World Models

Interactive video world models lose visual persistence beyond training horizons because temporal RoPE offsets become out-of-distribution; WorldTrace assigns compressed memory slots virtual in-distribution positions to restore addressability, boosting temporal consistency by 15.5% and episodic recall

Xindi Wu, Sven Elflein, James Lucas, Olga Russakovsky and 4 more

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 16 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 2026Department of Computer Science, Cornell Princeton Image restoration

QUTCC: Quantile Uncertainty Training and Conformal Calibration for Imaging Inverse Problems

QUTCC trains a U-Net for spatially adaptive quantile regression and calibrates tighter pixel-valid uncertainty intervals via non-linear conformal scaling for imaging inverse problems.

Cassandra T Ye, Shamus Li, Tyler King, Kristina Monakhova

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.

AI panel: 12 of 20 reviewers recommend it

lenient 5/5

medium 7/10

strict 0/5

91%Must read

?Must readVote to see the score

NeurIPS 2026Department of Computer Science, Princeton Carnegie Mellon U Toronto Meta Code generation

P$^{3}$: Joint Program-and-Proof Planning\\ for Verified Code Generation

P³ plans programs and proofs jointly from specifications before elaboration, outperforming sequential baselines by up to 11.2 points on verified generation benchmarks while reducing cost and time.

Zenan Li, Ziran Yang, Peiyang Song, Zhaoyu Li and 1 more

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

– ReadersNo votes yet

18/20 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: 18 of 20 reviewers recommend it

lenient 5/5

medium 10/10

strict 3/5

88%Must read

?Must readVote to see the score

NeurIPS 2026Eneco Princeton Optimization

Beyond the Half Approximation: Fair and Efficient Online Class Matching

Threshold-based algorithms achieve constant class envy-freeness and exceed 1/2 utilitarian welfare in online class matching, with near-matching upper bounds characterizing fairness costs.

Sander Borst, Max Springer

Paris Poster Session 1, Wed, Dec 9, 12:30 PM–2:30 PM, Paris Poster Hall · 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 4/5

medium 7/10

strict 4/5

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Princeton Jailbreaks & red teaming

The Geometry of Alignment Collapse: When Fine-Tuning Breaks Safety

Fine-tuning breaks safety via unstable geometric alignment subspaces, with alignment loss scaling quartically in training time via curvature-driven drift.

Max Springer, Chung Peng Lee, Bohdan Turbal, Blossom Metevier and 4 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.

AI panel: 11 of 20 reviewers recommend it

lenient 3/5

medium 7/10

strict 1/5

88%Must read

?Must readVote to see the score

NeurIPS 2026OralPrinceton MSL Mathematical reasoning

Escaping the Cognitive Well: Efficient Competition Math with Off-the-Shelf Models

An inference pipeline using off-the-shelf models and conjecture extraction with context detachment achieves state-of-the-art IMO-style math performance at much lower cost by escaping the Cognitive Well.

Xingyu Dang, Rohit Agarwal, Rodrigo Porto, Anirudh Goyal and 2 more

Atlanta Poster Session 6, Fri, Dec 11, 4:30 PM–7:30 PM, Hall C1 · 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 4/5

medium 9/10

strict 2/5

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Princeton U California, Santa Cruz Carnegie Mellon UC Berkeley Long-context modeling

Learning What to Remember: Test-Time Training via Context Distillation

TTCD uses a long-window teacher to supervise a short-window student's fast weights via hidden-state discrepancy, allocating limited memory to future-relevant context and outperforming existing long-context methods with minimal architectural changes.

Zixuan Wang, Xingyu Dang, Rui-Jie Zhu, Zixin Wen and 3 more

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

80%Must read

?Must readVote to see the score

NeurIPS 2026Princeton École Polytechnique Text-to-image

D2D: Detector-to-Differentiable Critic for Improved Numeracy in Text-to-Image Generation

D2D converts non-differentiable detectors into differentiable critics via custom activations to guide text-to-image numeracy, substantially improving object counting with minimal quality loss.

Nobline Yoo, Olga Russakovsky, Ye Zhu

Paris Poster Session 4, Thu, Dec 10, 5:30 PM–7:30 PM, Paris Poster Hall · Published 2026 · ▲ 3 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 7/10

strict 1/5

70%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Princeton The University of Hong Kong Northwestern Flow matching

A Theory on Flow Matching with Neural Networks

Flow matching with over-parameterized 2-layer ReLU networks achieves convergence, generalization, and sample-generation Wasserstein guarantees via multi-task representation bounds.

Yihan He, Qishuo Yin, Yuan Cao, Jianqing Fan and 1 more

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 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 2/5

medium 2/10

strict 1/5

89%Must read

?Must readVote to see the score

NeurIPS 2026Duke Princeton Apple Massachusetts Institute of Techn Stanford Vision-language models

LensVLM: Selective Context Expansion for Compressed Visual Representation of Text

LensVLM lets VLMs scan compressed rendered text and selectively expand only relevant regions via learned tools, maintaining near-full accuracy at 4.3x compression and outperforming baselines up to 10.1x across text QA benchmarks.

Roy Xie, Dan Friedman, Donghan Yu, Bowen Pan and 6 more

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 7 on Hugging Face

– 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 9/10

strict 2/5

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Princeton Massachusetts Institute of Techn Stanford MIT Purdue Multi-agent LLM systems

Improving the Efficiency of Language Agent Teams with Adaptive Task Graphs

LATTE coordinates LLM teams via shared evolving task graphs that reduce tokens, time, and conflicts while matching or exceeding baseline accuracy.

Elizabeth Mieczkowski, Alexander Ku, Tiwalayo Eisape, Dilip Arumugam and 4 more

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · Code ★ 6

– 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 2026Princeton Anthropic New York New York University; Anthropic MIT LLM evaluation & benchmarks

DiscoverPhysics: Benchmarking LLMs for out-of-the-box scientific thinking

DiscoverPhysics benchmarks LLM agents on simulated worlds with non-standard physics, finding frontier models pass only half and fail at uncovering latent structure.

Lindsay Smith, Matt Sampson, Siddharth Mishra-Sharma, Peter Melchior and 3 more

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

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 4/5

medium 9/10

strict 4/5

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Harvard University, Harvard Princeton Meta/Mila U California Los Angeles Vision-language models

Binding Visual Features Point by Point

Pointing via text induces internal visual search routines that eliminate binding errors, enabling compositional generalization and solving vision-language binding via serial processing.

Udith Haputhanthri, Declan Campbell, Rim Assouel, Jonathan D Cohen and 1 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 6/10

strict 0/5

88%Must read

?Must readVote to see the score

NeurIPS 2026Princeton Princeton LLM reasoning & chain of thought

Rethinking On-Policy Self-Distillation for Thinking Models

Privileged self-distillation degrades thinking models by suppressing reasoning forks and self-correction tokens, reducing long-rollout accuracy by up to 17%.

Simran Kaur, Narutatsu Ri, Yinghui He, Liam Fowl and 1 more

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

AI panel: 15 of 20 reviewers recommend it

lenient 3/5

medium 9/10

strict 3/5

86%Must read

?Must readVote to see the score