Ax Florin Gogianu, Adrian Catalin Lutu, Razvan Pascanu 5/11/2026

Revisiting Adam for Streaming Reinforcement Learning

Adam optimizer revisited for streaming reinforcement learning without replay buffers, enabling online policy updates from continuous interactions.

Ax Rachel Ma, Dylan Hadfield-Menell, Kristjan Greenewald 5/11/2026

Distributional Process Reward Models: Calibrated Prediction of Future Rewards via Conditional Optimal Transport

Calibration method for Process Reward Models using conditional optimal transport to improve inference-time scaling in LLM reasoning.

Ax Naihe Feng, Yi Sui, Shiyi Hou, Ga Wu, Jesse C. Cresswell 5/11/2026

Conformal Agent Error Attribution

Error attribution framework for multi-agent LLM systems using conformal prediction with distribution-free coverage guarantees.

Ax Quentin Berthet, Yu-Han Wu, Clement Crepy, Romuald Elie, Klaus Greff, Michael Eli Sander 5/11/2026

MIND: Monge Inception Distance for Generative Models Evaluation

Monge Inception Distance metric for evaluating generative models using sliced Wasserstein distance, addressing FID limitations.

Ax Debolina Halder Lina, Arlei Silva 5/11/2026

From Model to Data (M2D): Shifting Complexity from GNNs to Graphs for Transparent Graph Learning

Model-to-Data framework improving transparency and explainability of Graph Neural Networks by shifting complexity from models to data representations.

Ax Ilan Doron-Arad, Idan Mehalel, Elchanan Mossel 5/11/2026

A Theory of Online Learning with Autoregressive Chain-of-Thought Reasoning

PAC learning theory for autoregressive chain-of-thought reasoning in LLMs, extending online learning analysis to token generation mechanisms.

Ax Xiwen Chen, Wenhui Zhu, Songzhu Zheng, Kashif Rasul, Yueyue Deng, Huayu Li 5/11/2026

SHARP: A Self-Evolving Human-Auditable Rubric Policy for Financial Trading Agents

Self-evolving trading agent using interpretable rubric policy with bounded prompt optimization for financial markets.

Ax Aditya Ranganath, Mukesh Singhal 5/11/2026

A Unified Measure-Theoretic View of Diffusion, Score-Based, and Flow Matching Generative Models

Unified measure-theoretic framework showing diffusion, score-based, and flow matching as instances of vector field learning.

Ax Rui Zhu, Weiheng Bai, Qiushi Wu, Yang Ren, Haixu Tang, Yuchu Liu 5/11/2026

How to Compress KV Cache in RL Post-Training? Shadow Mask Distillation for Memory-Efficient Alignment

Shadow Mask Distillation technique compressing KV cache during RL post-training of LLMs for memory efficiency.

Ax Ishani Mondal, Shweta Bhardwaj 5/11/2026

Benchmarked Yet Not Measured -- Generative AI Should be Evaluated Against Real-World Utility

Analysis of benchmark-utility gap in generative AI across 28 real-world deployments, identifying evaluation failures.

Ax Pengrun Huang, Kamalika Chaudhuri, Yu-Xiang Wang 5/11/2026

Dataset Watermarking for Closed LLMs with Provable Detection

Dataset watermarking technique for closed LLMs enabling provable detection of proprietary or benchmark data usage.

Ax Fred Zhangzhi Peng, Alexis Fox, Anru R. Zhang, Alexander Tong 5/11/2026

Don't Retrain, Align: Adapting Autoregressive LMs to Diffusion LMs via Representation Alignment

Method to adapt autoregressive LMs to diffusion LMs via representation alignment without retraining.

Ax Shang-Ling Hsu, Mark Tenzer, Cyrus Shahabi, Khurram Shafique 5/11/2026

TraXion: Rethinking Pre-training Frameworks for Mobility and Beyond

TraXion proposes new pre-training framework for mobility data reflecting structural properties of human trajectories rather than importing language modeling objectives.

Ax Ziming Li, Jiatan Huang, Xiaoguang Guo, Guilin Wang, Chuxu Zhang 5/11/2026

Same Signal, Opposite Meaning: Direction-Informed Adaptive Learning for LLM Agents

Direction-informed adaptive test-time compute for LLM agents uses directional signal interpretation to decide when additional computation improves performance.

Ax Chacha Chen, Matthew J\"orke, Adam Goli\'nski, Masha Fedzechkina, Guillermo Sapiro, Sinead Williamson, Nicholas Foti 5/11/2026

LLMs are not (consistently) Bayesian: Quantifying internal (in)consistencies of LLMs' probabilistic beliefs

Studies internal inconsistencies in LLM probabilistic reasoning, analyzing whether LLMs update beliefs consistently with Bayesian principles as evidence changes.

Ax Fan Xu, Yuan Gao, Kun Wang, Rui Su, Fenghua Ling, Hao Wu, Wanli Ouyang 5/11/2026

Tyche: One Step Flow for Efficient Probabilistic Weather Forecasting

Tyche efficient probabilistic weather forecasting using one-step flow models, reducing inference cost compared to diffusion-based ensemble methods.

Ax Eshed Gal, Uri Ascher, Eldad Haber 5/11/2026

Target-Aware Data Augmentation for SAT Prediction

Target-aware data augmentation for SAT prediction reduces labeling costs by improving learning-based solvers for NP-hard Boolean satisfiability problems.

Ax Sepideh Avizeh, Tushin Mallick, Alina Oprea, Cristina Nita-Rotaru, Reihaneh Safavi-Naini 5/11/2026

MAGIQ: A Post-Quantum Multi-Agentic AI Governance System with Provable Security

MAGIQ proposes multi-agentic AI governance system with post-quantum cryptography for secure agent communication and accountability.

Ax Giansalvo Cirrincione, Adriano Fagiolini 5/11/2026

Learned Lyapunov Shielding for Adaptive Control

Learned Lyapunov shielding augments adaptive control with learned quadratic Lyapunov functions and physics-informed neural networks for safety filtering.

Ax Teo Susnjak 5/11/2026

A Reproducible Optimisation Protocol for Calibrating Prompt-Based Large Language Model Workflows in Evidence Synthesis

Reproducible calibration workflow for prompt-based LLMs in evidence synthesis tasks, separating task rules from mutable prompt harness with explicit metrics.

Ax James Fiedler 5/11/2026

Bias and Uncertainty in LLM-as-a-Judge Estimation

Analyzes systematic bias in LLM-as-a-Judge evaluation, proposing estimators to correct bias and improve calibration stability for model comparisons.

Ax Shivaram Subramanian, Zhengliang Xue, Markus Ettl, Yingdong Lu, Jayant Kalagnanam 5/11/2026

Causal-Aware Foundation-Model for Bilevel Optimization in Discrete Choice Settings

C3PO network applies causal-aware foundation models to bilevel optimization for dynamic pricing and assortment selection in discrete choice settings.

Ax Steven Song, Sahil Sethi, Brett Beaulieu-Jones, Robert L. Grossman 5/11/2026

ProtoSSL: Interpretable Prototype Learning from Unlabeled Time-Series Data

ProtoSSL enables interpretable time-series prediction through prototype learning from unlabeled data, providing case-based explanations.

Ax Chenhui Xu, Ziyue Bai, Fuxun Yu, Heng Huang, Jinjun Xiong 5/11/2026

Rollback-Free Stable Brick Structures Generation

RL-based approach for generating physically stable brick structures without external simulators, using learned constraint satisfaction during generation.

Ax Victor Livernoche, Jie Zan, Reihaneh Rabbany 5/11/2026

Kurtosis-Guided Denoising Score Matching for Tabular Anomaly Detection

Kurtosis-guided denoising score matching method for detecting anomalies in tabular data by learning score functions from noise-corrupted samples.

Ax Di Wu, Chengshuai Shi, Jing Yang, Cong Shen 5/11/2026

$f$-Divergence Regularized RLHF: Two Tales of Sampling and Unified Analyses

Unified theoretical analysis of f-divergence regularization in RLHF for LLM post-training, exploring alternatives to reverse KL divergence.

Ax Jonathn Chang, Arya Datla, Ziv Goldfeld 5/11/2026

PLOT: Progressive Localization via Optimal Transport in Neural Causal Abstraction

PLOT advances causal abstraction for neural network interpretability using optimal transport for efficient localization of relevant neural sites.

Ax Ahmed K. Kadhim, Lei Jiao, Rishad Shafik, Ole-Christoffer Granmo, Mayur Kishor Shende 5/11/2026

FastOmniTMAE: Parallel Clause Learning for Scalable and Hardware-Efficient Tsetlin Embeddings

FastOmniTMAE proposes parallel clause learning for efficient Tsetlin Machine embeddings in NLP, offering interpretable alternative to BERT and Word2Vec.

Ax Federico Echenique, Alireza Fallah, Baihe Huang, Michael I. Jordan 5/11/2026

Response Time Enhances Alignment with Heterogeneous Preferences

Framework using response time alongside binary choice data to align LLMs with heterogeneous user preferences without pooling feedback.

Ax Yonatan Slutzky, Yotam Alexander, Tomer Slor, Yoav Nagel, Nadav Cohen 5/11/2026

Why Does Agentic Safety Fail to Generalize Across Tasks?

Analysis of why safety measures in multi-task AI agents fail to generalize despite successful task execution generalization.

Ax Anupama Sridhar, Alexander Johansen 5/11/2026

Echo: KV-Cache-Free Associative Recall with Spectral Koopman Operators

Echo: KV-cache-free method using spectral Koopman operators for associative recall in long-context reasoning and tool-calling.

Ax Tianxin Zhou, Xiang Li, Haibing Lu 5/11/2026

Inductive Power Grid Cascading Failure Analysis with GRU-Gated Graph Attention

GRU-gated Graph Attention Network for identifying vulnerable transmission lines and predicting cascading failures in power grids.

Ax Da Long, Lingyi Fu, Diya Michelle Rao, Jasmine Ruales Carrera, Yang Bai, Shandian Zhe 5/11/2026

Dual-Agent Co-Training for Health Coaching via Implicit Adversarial Preference Optimization

Dual-agent framework using implicit adversarial preference optimization to improve AI health coaches based on motivational interviewing.

Ax Mahdi Erfanian, Nelson Daniel Troncoso, Aashna Garg, Amabel Gale, Xiaoyu Liu, Pareesa Ameneh Golnari, Shengyu Fu 5/11/2026

Delulu: A Verified Multi-Lingual Benchmark for Code Hallucination Detection in Fill-in-the-Middle Tasks

Delulu: Verified multi-lingual benchmark of 1,951 code hallucination samples in fill-in-the-middle tasks across 7 languages.

Ax Montaser Mohammedalamen, Kevin Roice, Reginald McLean, Alyssa Lefaivre \v{S}kopac 5/11/2026

A Systematic Investigation of The RL-Jailbreaker in LLMs

Mechanistic study of how RL-based adversarial attacks successfully jailbreak LLMs through multi-step sequential optimization.

Ax Aditya Sai Ellendula, Yi Wang, Chandrajit Bajaj 5/11/2026

Learning Material-Aware Hamiltonian Risk Fields for Safe Navigation

Port-Hamiltonian approach to risk-aware navigation policies that adapt evasive maneuvers based on local scene context.

Ax Minghao Yan, Bo Peng, Benjamin Coleman, Ziqi Chen, Zhouhang Xie, Shuo Chen, Zhankui He, Noveen Sachdeva, Weili Wang, Ed H. Chi, Shivaram Venkataraman, Wang-Cheng Kang, Derek Zhiyuan Cheng, Beidou Wang 5/11/2026

PACEvolve++: Improving Test-time Learning for Evolutionary Search Agents

PACEevolve++: Reinforcement learning framework enabling test-time policy adaptation for LLM-driven evolutionary search agents.

Ax Yi He, Xingyu Zhou 5/11/2026

Towards Differentially Private Reinforcement Learning with General Function Approximation

Theoretical framework for differentially private reinforcement learning with general function approximation beyond tabular/linear settings.

Ax Jiamin Xu, Jacqueline Maasch, Kyra Gan 5/11/2026

Integrating Causal DAGs in Deep RL: Activating Minimal Markovian States with Multi-Order Exposure

Method for discovering minimal Markovian states from causal DAGs in reinforcement learning without assuming states are pre-provided.

Ax Pingbang Hu, Xueshen Liu, Z. Morley Mao, Jiaqi W. Ma 5/11/2026

Dr. Post-Training: A Data Regularization Perspective on LLM Post-Training

Dr. Post-Training framework reconceptualizes data selection in LLM fine-tuning as regularization to prevent overfitting on scarce target data.

Ax Rui Cai, Weijie Jacky Mo, Xiaofei Wen, Qiyao Ma, Wenhui Zhu, Xiwen Chen, Muhao Chen, Zhe Zhao 5/11/2026

ModelLens: Finding the Best for Your Task from Myriads of Models

Framework for selecting best pretrained model for new tasks from thousands of open-source models using transferability estimation without expensive evaluation.

Ax Zekun Wang, Anant Gupta, Tianyi Zhu, Christopher J. MacLellan 5/11/2026

Test-Time Compositional Generalization in Diffusion Models via Concept Discovery

Approach for discovering and composing learned concepts from diffusion model score functions at test time for compositional generation.

Ax Majid Molaei, Gabor Paczolay, Matteo Papini, Alberto Maria Metelli, Marcello Restelli 5/11/2026

Actor-Critic with Active Importance Sampling

Actor-critic algorithm optimizing behavior policy via importance sampling to reduce variance in policy gradient estimation.

Ax Hayden Helm, Ben Johnson, Carey Priebe 5/11/2026

Query-efficient model evaluation using cached responses

Method for efficient model evaluation using cached responses from previously-evaluated models to reduce queries needed for benchmark assessment.

Ax Lucas Monteiro Paes, Natalie Mackraz, Barry-John Theobald, Federico Danieli 5/11/2026

Theoretical Limits of Language Model Alignment

Theoretical analysis of fundamental limits on reward improvement for LLM alignment via RL and best-of-N selection methods.

Ax Hao Chen, Chendi Qian, Christopher Morris, Andrea Lodi, Can Li 5/11/2026

Solving Max-Cut to Global Optimality via Feasibility-Preserving Graph Neural Networks

Graph neural network approach for solving max-cut combinatorial optimization within branch-and-bound using learned semidefinite relaxations.

Ax Tao Wang, Shuo Li, Yan Sun, Dongsheng Ding, Edgar Dobriban 5/11/2026

Where to Spend Rollouts: Hit-Utility Optimal Rollout Allocation for Group-Based RLVR

Optimal rollout allocation strategy for group-based RLVR improving LLM reasoning by dynamically distributing compute based on prompt saturation.

Ax Minseok Kim, Yeongjong Kim, Namkyeong Cho, Yeoneung Kim 5/11/2026

Stabilized neural Hamilton--Jacobi--Bellman solvers: Error analysis and applications in model-based reinforcement learning

Error analysis and applications of neural solvers for Hamilton-Jacobi-Bellman equations in continuous-time model-based reinforcement learning.

Ax Zixuan Xie, Xinyu Liu, Rohan Chandra, Shangtong Zhang 5/11/2026

Convergence and Emergence of In-Context Reinforcement Learning with Chain of Thought

Theoretical analysis of in-context reinforcement learning with chain-of-thought, explaining convergence and emergence of adaptation capabilities at inference time.

Ax Zhanqi Zhang, Hua-Dong Xiong, Robert C. Wilson, Mikio Aoi, Marcelo G. Mattar, Li Ji-An 5/11/2026

The Position Curse: LLMs Struggle to Locate the Last Few Items in a List

Empirical study revealing LLMs fail at retrieving last items in short lists despite strong few-shot performance, characterizing the 'Position Curse' failure mode.