Good Papers: read the good ones

Symmetry in variational inference forces approximate minimizers to recover target statistics under misspecification, unifying prior results and yielding new directional guarantees.

Daniel Marks, Dario Paccagnan, Mark van der Wilk

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.

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U Oxford ETH Zurich Oxofrd, University of Oxford Optimization

Time-Sensitive Anytime-Valid Testing

A time-sensitive testing-by-betting framework favors early rejection via time-weighted rewards, yielding Bellman-optimal e-processes and an exponential-decay-optimal criterion recovering classical growth-rate optimality at large scales.

Eugenio Clerico, Tobias Wegel, Iskander Azangulov, Patrick Rebeschini

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.

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.

74%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026MATS EPFL ETH Zurich U Oxford Probing & representation analysis

Probing Persona-Dependent Preferences in Language Models

Linear probes on LLM residual streams identify a shared preference vector tracking pairwise choices across personas, with cross-persona transfer and causal steering.

Oscar Gilg, Pierre Beckmann, Daniel Paleka, Patrick Butlin

Paris Poster Session 5, Fri, Dec 11, 11:30 AM–1:30 PM, Paris Poster Hall · 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.

83%Must read

?Must readVote to see the score

NeurIPS 2026Worcester Polytechnic Institute Apollo Research U Oxford Independent Google DeepMind AI oversight & deception

Evaluating and Understanding Scheming Propensity in LLM Agents

Realistic agent settings show minimal scheming despite high incentives, with model-organism scheming brittle to tool removal and oversight.

Mia Hopman, Jannes Elstner, Maria Avramidou, Amritanshu Prasad and 1 more

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

83%Must read

?Must readVote to see the score

NeurIPS 2026U British Columbia Tübingen AI Center U Oxford InstaDeep Research U Pennsylvania Controllable generation

Search at the Cost of Sampling: Nearly-Instant Latent Space Bayesian Optimization

Exploiting spherical latent geometry yields nearly closed-form Bayesian optimization with 100x speedups and matching performance in generative discovery pipelines.

Donney Fan, Colin Doumont, Aleksandra Kalisz, Paul Duckworth and 3 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 2026Nanyang Technological Mila - Quebec Artificial Intelli U Oxford U Montreal, Mila Deep RL

Beyond One-Size-Fits-All: Diagnosis-Driven Online Reinforcement Learning with Offline Priors

Diagnosis-driven tension management adapts online RL to deployment-specific prior validity shifts, rejecting universal benchmarks for flexible, evidence-guided optimization.

Guozheng Ma, Lu Li, Zilin Wang, Pierre-Luc Bacon 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.

71%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U Oxford Max Planck Institute for Multidi Institue for Advanced Study, Pri Max PlancInstitute for Multidisc Oxford Graph generation

Permutation-Invariant Spectral Learning via Dyson Diffusion

Dyson Diffusion Model uses Dyson Brownian motion to shift inductive bias to diffusion dynamics, yielding permutation-invariant spectral learning that improves graph generation.

Tassilo Schwarz, Cai Dieball, Constantin Kogler, Renaud Lambiotte and 3 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.

AI panel: 6 of 20 reviewers recommend it

lenient 2/5

medium 4/10

strict 0/5

91%Must read

?Must readVote to see the score

NeurIPS 2026U Edinburgh U Oxford Google DeepMind Toyota Research Institute University College London, Googl Multi-agent LLM systems

Benchmarking Open-Ended Multi-Agent Coordination in Language Agents

Alem benchmarks open-ended multi-agent coordination for language agents, showing frontier LLMs average ~6% returns and individual competence does not imply coordination competence.

Kale-ab Tessera, Andras Szecsenyi, Cameron Barker, Alexander Rutherford and 6 more

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 7 on Hugging Face · Code ★ 51

– 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

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Technion U Oxford Technion---Israel Institute of T AI oversight & deception

Selective Safety Steering via Value-Filtered Decoding

Value-filtered decoding selectively steers LLM generation using a value-based safety criterion with explicit false-intervention bounds, improving safety-utility trade-offs over baselines.

Bat-Sheva Einbinder, Hen Davidov, Yee Whye Teh, Yarin Gal and 1 more

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

80%Must read

?Must readVote to see the score

NeurIPS 2026U Oxford New York University Of Oxford Multi-agent RL

Modelling Opinion Dynamics at Scale with Deep MARL

Deep MARL scales opinion dynamics to 1000 agents, finding high conformity in large networks reduces accuracy and promotes dishonesty, unlike small groups, revealing a mismatch with modern media.

Lukas Seier, Brandon Kaplowitz, Sebastian Towers, Richard M Bailey and 1 more

Sydney Poster Session 5, Thu, Dec 10, 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.

80%Must read

?Must readVote to see the score

NeurIPS 2026U Oxford Meta FAIR Carnegie Mellon U Washington FAIR @ Meta Meta Superintelligence Labs Test-time compute

Social Choice Foundations for Simulation-Augmented Generation

SAGE formalizes efficient inference-time viewpoint simulation via metric proportional justified representation, proving small simulated pools and dynamic routing preserve approximate proportional representation for contentious queries.

Sonja Kraiczy, Smitha Milli, Ratip Emin Berker, Avinandan Bose and 5 more

Sydney Poster Session 5, Thu, Dec 10, 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.

83%Must read

?Must readVote to see the score

NeurIPS 2026U Oxford Skyral Department of Computer Science Department of Computer Science U Univ of Oxford Bandits & online learning

Bayesian Decision Making around Experts

In Bayesian multi-armed bandits, pretraining on expert data tightens regret bounds by mutual information with the optimal action, while an information-directed rule selects data sources maximizing immediate information gain, and trust inference safeguards against ineffective or compromised experts.

Daniel Jarne Ornia, Joel Dyer, Nicholas Bishop, Anisoara Calinescu and 1 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.

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 2026U Oxford Boston Deep learning theory

Convergence Analysis of Newton's Method for Neural Networks in the Overparameterized Limit

Regularized Newton training of overparameterized neural networks converges to a deterministic NNTK limit with exponentially fast uniform convergence across all frequencies, avoiding gradient descent's spectral bias.

Konstantin Riedl, Justin Sirignano, Konstantinos Spiliopoulos

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.

72%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Friedrich-Alexander Universität U Wuerzburg U Oxford Technical University of Munich Deep learning theory

Quantifying Concentration Phenomena of Mean-Field Transformers in the Low-Temperature Regime

Mean-field transformers exhibit rapid token distribution concentration onto projection-driven limits with explicit Wasserstein bounds scaling in inverse temperature β and time t.

Albert Alcalde, Leon Bungert, Konstantin Riedl, Tim Roith

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

71%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U Oxford Inria Paris / ENS U Copenhagen Diffusion models

Metropolis-Adjusted Diffusion Models

Metropolis-adjusted Langevin correctors using score-based acceptance probabilities and a two-coin Bernoulli factory reduce diffusion model sampling bias and improve FID.

Kevin H. Lam, Tyler Farghly, Christopher Williams, Jun Yang and 2 more

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

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026City University of Hong Kong Imperial College London City University of Hong Kong U Oxford AstraZeneca Time series foundation models

Extending Pretrained 10-Second ECG Foundation Models to Longer Horizons

A parameter-efficient plugin extends frozen 10-second ECG foundation models to long, variable-length recordings via compatible long-sequence processing and semantically informed temporal modeling, outperforming sliding-window and pooling baselines.

Wei Tang, Jinpei Han, Kangning Cui, Mattia Carletti and 9 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 5/5

medium 6/10

strict 0/5

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U Oxford Bayesian & probabilistic methods

Asymptotically Log-Optimal Bayes-Assisted Confidence Sequences for Bounded Means

A Bayes-assisted framework adaptively builds confidence sequences via predictive expected log-growth to achieve asymptotic log-optimality and narrower widths.

Valentin Kilian, Stefano Cortinovis, Francois Caron

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 3/5

medium 6/10

strict 2/5

91%Must read

?Must readVote to see the score

NeurIPS 2026U Oxford U Toronto IDEAS NCBR Sp. z o. o. Address: MATS national university of singaore,Hallucination & factuality

Negation Neglect: When models fail to learn negations in training

Fine-tuning LLMs on documents that flag claims as false makes them believe those claims, with belief rates jumping from 2.5% to 88.6%, though local negation phrasing largely prevents it.

Harry Mayne, Lev McKinney, Jan Dubiński, Adam Karvonen and 2 more

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

71%Highly rated

?Highly ratedVote to see the score

ICML 2026WorkshopU Oxford King's College London, Univ Controllable generation

A Mean-Field Framework for Inference-Time Distributional Control of Diffusion Models

A mean-field framework formulates inference-time diffusion control via weighted interacting particles to target distribution-level rewards with theoretical guarantees.

Samuel Howard, Nikolas Nüsken

Sydney Poster Session 2, Tue, Dec 8, 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.

AI panel: 7 of 20 reviewers recommend it

lenient 3/5

medium 3/10

strict 1/5

71%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Stanford U Oxford UCL & FAIR Meta Google DeepMind LLM reasoning & chain of thought

Asking the Right Questions: Improving Reasoning with Generated Stepping Stones

ARQ introduces a question generator that produces transferable intermediate stepping stones, improving reasoning LLM performance via fine-tuning on synthetic data.

Hengyuan Hu, Tingchen Fu, Minqi Jiang, Alexander Miller and 2 more

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 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.

AI panel: 6 of 20 reviewers recommend it

lenient 4/5

medium 2/10

strict 0/5

71%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U Oxford Mila / Université de Montréal Google DeepMind / U. de Montreal Mila, U. Montreal USI Mechanistic interpretability

A Mechanistic Analysis of Looped Reasoning Language Models

Looped reasoning models converge to cyclic fixed points that stabilize attention and repeat feedforward inference stages iteratively, with recurrence size and normalization affecting stability.

Hugh Blayney, Alvaro Arroyo, Johan Obando Ceron, Pablo Samuel Castro and 3 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

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 3/5

medium 3/10

strict 0/5

91%Must read

?Must readVote to see the score

NeurIPS 2026U Oxford U Washington University College London, Unive Pepperdine Imperial College London Adversarial robustness

MedMisBench: Measuring Epistemic Resilience of LLMs Under Misleading Medical Context

MedMisBench reveals LLM medical accuracy collapses from 71% to 38% under misleading context, exposing a critical evaluation blind spot around epistemic resilience.

Hongjian Zhou, Xinyu Zou, Jinge Wu, Sean Wu and 18 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.

AI panel: 17 of 20 reviewers recommend it

lenient 5/5

medium 8/10

strict 4/5

83%Must read

?Must readVote to see the score

NeurIPS 2026U Oxford Retrieval-augmented generation

Beyond Raw Context Transfer: Representation-based Federated Retrieval-Augmented Generation

FedRepRAG is a federated RAG framework that exchanges only compact latent representations across clients to reduce inference overhead, outperforming local retrieval baselines on decentralized VQA and QA tasks.

Can Peng, Yu Liu, Yingyu Yang, Anjie Le and 3 more

Paris Poster Session 6, Fri, Dec 11, 2:30 PM–4:30 PM, Paris Poster Hall · 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 8/10

strict 0/5

Show 20 more papers