Ax Ruotong Sun, Ermin Wei 5/8/2026

Pro-KLShampoo: Projected KL-Shampoo with Whitening Recovered by Orthogonalization

Combines Kronecker-factored preconditioning with gradient momentum orthogonalization for LLM pre-training optimization.

Ax Elizabeth Mieczkowski, Alexander Ku, Tiwalayo Eisape, Dilip Arumugam, John Matters, Katherine M. Collins, Ilia Sucholutsky, Thomas L. Griffiths 5/8/2026

Improving the Efficiency of Language Agent Teams with Adaptive Task Graphs

Proposes adaptive task graphs to improve coordination efficiency in LLM agent teams, reducing error propagation and resource waste.

Ax Florian A. D. Burnat, Brittany I. Davidson 5/8/2026

Measuring Evaluation-Context Divergence in Open-Weight LLMs: A Paired-Prompt Protocol with Pilot Evidence of Alignment-Pipeline-Specific Heterogeneity

Measures evaluation-context divergence in open-weight LLMs using paired prompts to detect alignment pipeline-specific behavioral differences.

Ax Siraaj Akhtar, Saad Khan, Simon Parkinson 5/8/2026

Fine-Tuning Small Language Models for Solution-Oriented Windows Event Log Analysis

Fine-tunes small language models for Windows event log analysis with actionable remediation, addressing privacy and resource constraints of LLMs.

Ax Thomas B\"omer, Bastian Amberg, Max Disselnmeyer, Anne Meyer 5/8/2026

CoupleEvo: Evolving Heuristics for Coupled Optimization Problems Using Large Language Models

Uses LLMs with evolutionary algorithms to design heuristics for coupled optimization problems across multiple interdependent subproblems.

Ax Xuening Wu, Yanlan Kang, Qianya Xu, Kexuan Xie, Jiaqi Mi, Honggang Wang, Yubin Liu, Zeping Chen 5/8/2026

Human-AI Co-Evolution and Epistemic Collapse: A Dynamical Systems Perspective

Analysis of feedback loops between humans and LLMs as coupled dynamical systems, studying knowledge collapse from recursive AI-generated training.

Ax Dylan Bouchard 5/8/2026

Is Escalation Worth It? A Decision-Theoretic Characterization of LLM Cascades

Decision-theoretic framework for LLM cascades analyzing optimal deferral thresholds between models to balance cost and quality tradeoffs.

Ax Yifan Tang, Qiquan Wang, In\'es Garc\'ia-Redondo, Anthea Monod 5/8/2026

Topological Signatures of Grokking

Topological analysis of grokking phenomenon in neural networks using persistent homology on embedding matrices.

Ax Yuan Du, Mitchel Hill, HanQin Cai 5/8/2026

Memory Efficient Full-gradient Attacks (MEFA) Framework for Adversarial Defense Evaluations

Memory-efficient gradient computation method for evaluating adversarial robustness of diffusion and Langevin-based defense mechanisms.

Ax Xin Peng, Ang Gao 5/8/2026

Flow Matching with Arbitrary Auxiliary Paths

Generative modeling framework extending flow matching with arbitrary auxiliary distributions for flexible trajectory generation.

Ax Paulo Mario P. Medina, Jose Marie Antonio Mi\~noza, Sebastian C. Iba\~nez 5/8/2026

eXplaining to Learn (eX2L): Regularization Using Contrastive Visual Explanation Pairs for Distribution Shifts

Framework using visual explanations as regularization to improve model robustness under distribution shifts, addressing spurious correlations interpretably.

Ax Tao Liu, Hao Yan, Mengting Chen, Taihang Hu, Zhengrong Yue, Zihao Pan, Jinsong Lan, Xiaoyong Zhu, Ming-Ming Cheng, Bo Zheng, Yaxing Wang 5/8/2026

Continuous-Time Distribution Matching for Few-Step Diffusion Distillation

Continuous-time distribution matching for accelerating diffusion model distillation with sparse supervision.

Ax Nan Jia, Haojin Yang, Xing Ma, Jiesong Lian, Shuailiang Zhang, Weipeng Zhang, Ke Zeng, Xunliang Cai, Zequn Sun 5/8/2026

Asymmetric On-Policy Distillation: Bridging Exploitation and Imitation at the Token Level

On-policy distillation method addressing token-level learning between exploitation and imitation with improved gradient properties.

Ax Kartik Tandon, Julian Gould, Tanishq Bhatia, Francesca Dominici, Alejandro Ribeiro, Claudio Battiloro 5/8/2026

Consistent Geometric Deep Learning via Hilbert Bundles and Cellular Sheaves

Convolutional learning framework for infinite-dimensional signals on manifolds using Hilbert bundles and cellular sheaves.

Ax Qingjun Zhang 5/8/2026

E = T*H/(O+B): A Dimensionless Control Parameter for Mixture-of-Experts Ecology

Dimensionless control parameter predicting Mixture-of-Experts model collapse, unifying expert ecology across vision and language.

Ax Francesco Dente, Dario Satriani, Paolo Papotti 5/8/2026

Constraint Decay: The Fragility of LLM Agents in Backend Code Generation

Study of LLM agents failing to maintain structural constraints in backend code generation, analyzing non-functional requirement violations.

Ax Maresa Schr\"oder, Pascal Janetzky, Michael Klar, Stefan Feuerriegel 5/8/2026

ORTHOBO: Orthogonal Bayesian Hyperparameter Optimization

Bayesian hyperparameter optimization addressing acquisition estimation noise through orthogonal sampling.

Ax Xiang Li, Nan Jiang 5/8/2026

Q-MMR: Off-Policy Evaluation via Recursive Reweighting and Moment Matching

Off-policy evaluation framework using recursive reweighting and moment matching for finite-horizon MDPs.

Ax Nii Osae Osae Dade, Tony Morri, Moinul Hossain Rahat, Sayandip Pal 5/8/2026

Litespark Inference on Consumer CPUs: Custom SIMD Kernels for Ternary Neural Networks

Custom SIMD kernels for running ternary neural networks on consumer CPUs, enabling LLM inference without GPUs.

Ax Zuyuan Zhang, Fei Xu Yu, Tian Lan 5/8/2026

Operator-Guided Invariance Learning for Continuous Reinforcement Learning

RL method for continuous spaces discovering value-preserving structures via operator-guided invariance learning.

Ax Murat Bilgehan Ertan, Xiaochen Zhu, Phuong Ha Nguyen, Marten van Dijk, Srinivas Devadas 5/8/2026

PACZero: PAC-Private Fine-Tuning of Language Models via Sign Quantization

PAC-private zeroth-order mechanism for fine-tuning LLMs via sign quantization with strong membership inference attack resistance.

Ax Amir Rezaei Balef, Mykhailo Koshil, Katharina Eggensperger 5/8/2026

Is One Layer Enough? Understanding Inference Dynamics in Tabular Foundation Models

Mechanistic study of inference dynamics in tabular transformer foundation models, analyzing layerwise prediction emergence.

Ax Haochen Cai, Xian Yu 5/8/2026

Learning to Cut: Reinforcement Learning for Benders Decomposition

RL framework for Benders decomposition that adaptively selects cuts using neural network policies for stochastic programming.

Ax Hao Ye, Jisheng Dang, Junfeng Fang, Bimei Wang, Yizhou Zhang, Ning Lv, Wencan Zhang, Hong Peng, Bin Hu, Tat-Seng Chua 5/8/2026

On the Implicit Reward Overfitting and the Low-rank Dynamics in RLVR

Study of implicit reward overfitting in reinforcement learning with verifiable rewards, analyzing low-rank dynamics in model reasoning.

Ax Hongcan Guo, Qinyu Zhao, Yian Zhao, Shen Nie, Rui Zhu, Qiushan Guo, Feng Wang, Tao Yang, Hengshuang Zhao, Guoqiang Wei, Yan Zeng 5/8/2026

Continuous Latent Diffusion Language Model

Hierarchical latent diffusion language model for text generation using non-autoregressive approach with improved efficiency and semantic modeling.

Ax Maria Ana Cardei, Matthew Landers, Afsaneh Doryab 5/8/2026

Coordination Matters: Evaluation of Cooperative Multi-Agent Reinforcement Learning

Coordination-aware evaluation metrics for cooperative multi-agent reinforcement learning with process-level diagnostics.

Ax Victor Daniel Gera 5/8/2026

Directional Consistency as a Complementary Optimization Signal: The GONO Framework

GONO optimization framework identifying decoupling between directional alignment and loss convergence in deep learning.

Ax Jyotirmaya Shivottam, Subhankar Mishra 5/8/2026

Towards Metric-Faithful Neural Graph Matching

Neural architecture for graph matching that estimates graph edit distance using GNNs with focus on encoder geometry.

Ax Shuyang Jiang, Nan Yu, Yiming Zhang, Zenghui Ding, Zhenyu Wu 5/8/2026

DINORANKCLIP: DINOv3 Distillation and Injection for Vision-Language Pretraining with High-Order Ranking Consistency

Vision-language pretraining method combining DINOv3 distillation with ranking consistency for improved CLIP models.

Ax Shuo Liu, Xinzichen Li, Christopher Amato 5/8/2026

Cross-Modal Navigation with Multi-Agent Reinforcement Learning

Multi-agent reinforcement learning approach for embodied navigation using specialized agents for different sensory modalities.

Ax Yiqiao Jin, Yiyang Wang, Lucheng Fu, Yijia Xiao, Yinyi Luo, Haoxin Liu, B. Aditya Prakash, Josiah Hester, Jindong Wang, Srijan Kumar 5/8/2026

UniSD: Towards a Unified Self-Distillation Framework for Large Language Models

Unified self-distillation framework for LLMs enabling adaptation without external teachers through self-generated trajectories and supervision.

Ax Isaac David, Arthur Gervais 5/8/2026

Patch2Vuln: Agentic Reconstruction of Vulnerabilities from Linux Distribution Binary Patches

Language model agent reconstructs security vulnerabilities from Linux binary patches using local binary evidence without source code access.

Ax Nithin Somasekharan, Rabi Pathak, Manushri Dhanakoti, Tingwen Zhang, Ling Yue, Andy Zhu, Shaowu Pan 5/8/2026

AI CFD Scientist: Toward Open-Ended Computational Fluid Dynamics Discovery with Physics-Aware AI Agents

Open-source AI agent for computational fluid dynamics discovery combining LLMs with physics simulators to automate scientific discovery loops.

Ax Siquan Li, Kaiqi Jiang, Jiacheng Sun, Tianyang Hu 5/8/2026

The Structural Origin of Attention Sink: Variance Discrepancy, Super Neurons, and Dimension Disparity

Mechanistic explanation of attention sink phenomenon in LLMs, tracing root causes to variance discrepancy and dimension disparity in self-attention.

Ax Hongyi Tao, Dingzhi Yu, Lijun Zhang 5/8/2026

When and Why SignSGD Outperforms SGD: A Theoretical Study Based on $\ell_1$-norm Lower Bounds

Theoretical analysis explaining when sign-based optimization algorithms like SignSGD outperform vanilla SGD for training large models.

Ax Apurva Gandhi, Satyaki Chakraborty, Xiangjun Wang, Aviral Kumar, Graham Neubig 5/8/2026

Recursive Agent Optimization

Reinforcement learning approach enabling agents to recursively spawn and delegate sub-tasks for divide-and-conquer inference-time scaling.

Ax Ronaldo Canizales, Divya Gopinath, Corina P\u{a}s\u{a}reanu, Ravi Mangal 5/8/2026

Concept-Based Abductive and Contrastive Explanations for Behaviors of Vision Models

Method for generating concept-based causal explanations of vision models using abductive and contrastive reasoning.

Ax Xiangyuan Xue, Yifan Zhou, Zidong Wang, Shengji Tang, Philip Torr, Wanli Ouyang, Lei Bai, Zhenfei Yin 5/8/2026

StraTA: Incentivizing Agentic Reinforcement Learning with Strategic Trajectory Abstraction

Framework for training LLM agents for long-horizon decision making using strategic trajectory abstraction to improve exploration and credit assignment.

Ax Hao Dong, Hongzhao Li, Shupan Li, Muhammad Haris Khan, Eleni Chatzi, Olga Fink 5/8/2026

Are We Making Progress in Multimodal Domain Generalization? A Comprehensive Benchmark Study

Comprehensive benchmark study on multimodal domain generalization methods, examining whether performance improvements reflect genuine algorithmic progress.

Ax Zeyu Yang, Qi Ma, Jason Chen, Anshumali Shrivastava 5/8/2026

Superintelligent Retrieval Agent: The Next Frontier of Information Retrieval

Retrieval-augmented agent system treating retrieval as active expert navigation rather than black-box queries for knowledge base interaction.

Ax Sushant Gautam, Finn Schwall, Annika Willoch Olstad, Fernando Vallecillos Ruiz, Birk Torpmann-Hagen, Sunniva Maria Stordal Bj{\o}rklund, Leon Moonen, Klas Pettersen, Michael A. Riegler 5/8/2026

When No Benchmark Exists: Validating Comparative LLM Safety Scoring Without Ground-Truth Labels

Framework for validating comparative LLM safety scores without labeled benchmarks, enabling safety evaluation across diverse languages and domains.

Ax Yuxing Liu, Jianyu Wang, Tong Zhang 5/8/2026

Optimizer-Model Consistency: Full Finetuning with the Same Optimizer as Pretraining Forgets Less

Study showing that LLM fine-tuning with the same optimizer as pretraining reduces catastrophic forgetting while maintaining performance on new tasks.

Ax Yuhang Lai, Jiazhan Feng, Yee Whye Teh, Ning Miao 5/8/2026

Verifier-Backed Hard Problem Generation for Mathematical Reasoning

Method for generating valid, challenging mathematical problems using LLM verifiers and self-play, enabling automated problem creation for LLM training.

Ax Borui Zhang, Bo Zhang, Bo Wang, Wenzhao Zheng, Yuhao Cheng, Liang Tang, Yiqiang Yan, Jie Zhou, Jiwen Lu 5/8/2026

BAMI: Training-Free Bias Mitigation in GUI Grounding

Training-free bias mitigation method for GUI grounding in agents, improving performance on complex screen interaction tasks.

Ax Peng Liu 5/8/2026

Multi-Modality Distillation via Learning the teacher's modality-level Gram Matrix

Knowledge distillation method for multi-modal models focusing on transferring modality-level relationships from teacher to student networks.

Ax Charlie Griffin, Louis Thomson, Buck Shlegeris, Alessandro Abate 5/8/2026

Games for AI Control: Models of Safety Evaluations of AI Deployment Protocols

Formal game-theoretic framework for safety evaluation of AI deployment protocols through red-teaming exercises.

Ax Rachel Ma, Jingyi Qu, Andreea Bobu, Dylan Hadfield-Menell 5/8/2026

Flexible Agent Alignment with Goal Inference from Open-Ended Dialog

Framework for aligning LLM-based agents with human preferences through goal inference from multi-turn dialogue, addressing challenges in collaborative agent interactions.

Ax Yidan Sun, Viktor Schlegel, Srinivasan Nandakumar, Iqra Zahid, Yuping Wu, Yulong Wu, Hao Li, Jie Zhang, Warren Del-Pinto, Goran Nenadic, Siew Kei Lam, Anil Anthony Bharath 5/8/2026

SynBench: A Benchmark for Differentially Private Text Generation

Benchmark for evaluating differentially private text generation methods using LLMs, enabling secure sharing of sensitive datasets across institutions.

Ax Sihan Hu, Xiansheng Cai, Yuan Huang, Zhiyuan Yao, Linfeng Zhang, Pan Zhang, Youjin Deng, Kun Chen 5/8/2026

Emergent Slow Thinking in LLMs as Inverse Tree Freezing

Research on how reinforcement learning enables LLMs to develop multi-step reasoning capabilities, using statistical physics framework to explain emergence of slow thinking.

Ax Claudio Fanconi, Nicol\'as Astorga, Mihaela van der Schaar 5/8/2026

Learning Reasoning Rewards from Expert Demonstrations with Inverse Reinforcement Learning

Learning reasoning reward functions for LLMs via inverse reinforcement learning from expert demonstrations, avoiding manual reward specification.