Ax Zhiyuan Xu, Joseph Gardiner, Sana Belguith, Lichao Wu 5/6/2026

RouteHijack: Routing-Aware Attack on Mixture-of-Experts LLMs

Adversarial attack on Mixture-of-Experts LLMs exploiting routing mechanisms to bypass safety alignment.

Ax Mohit Kumar, Somayeh Kargaran, Bernhard A. Moser, Manuela Gei{\ss} 5/6/2026

Kernel Affine Hull Machines for Compute-Efficient Query-Side Semantic Encoding

Kernel Affine Hull Machines enabling efficient query-side semantic encoding without repeated neural inference for transformer retrieval systems.

Ax Zhaoyuan Su, Olatunji Ruwase, Karthik Ganesan, Aurick Qiao, Samyam Rajbhandari, Juncheng Yang, Yue Cheng, Yuxiong He 5/6/2026

ZeRO-Prefill: Zero Redundancy Overheads in MoE Prefill Serving

ZeRO-Prefill optimization reducing distributed execution overhead in mixture-of-experts model serving for prefill-only discriminative tasks.

Ax Michael Chertkov 5/6/2026

Analytic Bridge Diffusions for Controlled Path Generation

Bridge diffusion method with closed-form solutions for score functions and drift fields enabling analytical controlled path generation without neural networks.

Ax Barbara Tarantino, Sun Kim, Yijingxiu Lu, Paolo Giudici 5/6/2026

ISAAC: Auditing Causal Reasoning in Deep Models for Drug-Target Interaction

ISAAC framework auditing causal reasoning in deep learning drug-target interaction models via intervention-based structural sensitivity probing.

Ax Kunvar Thaman 5/6/2026

Reward Hacking Benchmark: Measuring Exploits in LLM Agents with Tool Use

Benchmark for evaluating LLM agents with tool use detecting reward hacking exploits through multi-step task evaluation with naturalistic shortcuts.

Ax Yang Fu, Peng Qin, Liming Chen, Zihao Zhang, Hao Yu, Yifei Wang 5/6/2026

Joint Energy Management and Coordinated AIGC Workload Scheduling for Distributed Data Centers: A Diffusion-Aided Reward Shaping Approach

Diffusion-aided reward shaping approach for scheduling AIGC workloads across distributed data centers while minimizing energy costs.

Ax Xintan Zeng, Yongchao Liu, Yice Luo, Jiajun Zhen 5/6/2026

AutoRAGTuner: A Declarative Framework for Automatic Optimization of RAG Pipelines

AutoRAGTuner declarative framework automating RAG pipeline optimization through modular architecture and configuration-driven hyperparameter tuning.

Ax Ping Wang, Yan-Qi Du 5/6/2026

Finite-Size Gradient Transport in Large Language Model Pretraining: From Cascade Size to Intensive Transport Efficiency

Gradient transport analysis framework for understanding cascade efficiency and transport mechanisms during large language model pretraining.

Ax Ruiyang Qin, Qingzhuo Wang, Dongrui Liu, Qiang Li, Zhihua Wei, Wen Shen 5/6/2026

Multilingual Safety Alignment via Self-Distillation

Cross-lingual safeguard transfer framework improving multilingual safety alignment in LLMs through self-distillation from high-resource languages.

Ax Eitan Kosman, Gabriele Serussi, Chaim Basking 5/6/2026

Structured Diffusion Bridges: Inductive Bias for Denoising Diffusion Bridges

Diffusion bridge framework for unpaired modality translation with structured constraints to restrict solution space.

Ax Gideon Stein, Niklas Penzel, Tristan Piater, Joachim Denzler 5/6/2026

TCD-Arena: Assessing Robustness of Time Series Causal Discovery Methods Against Assumption Violations

TCD-Arena benchmark for evaluating robustness of time series causal discovery algorithms against violation of assumptions.

Ax Francesco Sovrano, Gabriele Dominici, Marc Langheinrich 5/6/2026

Neuron-Anchored Rule Extraction for Large Language Models via Contrastive Hierarchical Ablation

MechaRule framework extracting symbolic rules from LLMs by grounding rule extraction in neuron-level mechanistic interpretability via contrastive ablation.

Ax Sarvesh Patil, Mitsuhiko Nakamoto, Manan Agarwal, Shashwat Saxena, Jesse Zhang, Giri Anantharaman, Cleah Winston, Chaoyi Pan, Douglas Chen, Nai-Chieh Huang, Zeynep Temel, Oliver Kroemer, Sergey Levine, Abhishek Gupta, Hongkai Da, Paarth Shah, Max Simchowitz 5/6/2026

OGPO: Sample Efficient Full-Finetuning of Generative Control Policies

Sample-efficient fine-tuning algorithm for diffusion and flow-based robot control policies using off-policy critic networks and modified PPO.

Ax Adnan Ali, Jinlong Li, Syed Muhammad Israr, Ali Kashif Bashir 5/6/2026

Adaptive Negative Scheduling for Graph Contrastive Learning

Adaptive negative sampling scheduler for graph contrastive learning improving computational efficiency in self-supervised representation learning.

Ax Shubham Harkare, Arvind Yogesh Suresh Babu, Yash Kulkarni 5/6/2026

Attribution-Guided Masking for Robust Cross-Domain Sentiment Classification

Attribution-guided masking method for improving transformer sentiment classification transfer across domains by identifying spurious tokens.

Ax Zhecheng Sheng, Yongsen Tan, Xiruo Ding, Trevor Cohen, Serguei Pakhomov 5/6/2026

When Prompts Interact: Assessing Prompt Arithmetic for Deconfounding under Distribution Shift

Prompt arithmetic technique for improving LLM robustness under distribution shift without full fine-tuning via task arithmetic and prompt manipulation.

Ax Tianlu Lu, Asif Sijan, Thomas Noh, Huaijin Chen, Andrey A. Popov 5/6/2026

Pose Tracking with a Foundation Pose Model and an Ensemble Directional Kalman Filter

arXiv paper introducing Ensemble Directional Kalman Filter for pose tracking using unit-quaternion attitude representation beyond standard Kalman assumptions.

Ax Stephen J. Thomas 5/6/2026

Gated Subspace Inference for Transformer Acceleration

arXiv paper on Gated Subspace Inference accelerating transformer inference by exploiting low-rank activation manifolds with cached weights and per-token gating.

Ax Jingchu Gai, Laixi Shi 5/6/2026

Taming the Curses of Multiagency in Robust Markov Games with Large State Space through Linear Function Approximation

arXiv paper on distributionally robust Markov games for multi-agent RL addressing robustness and data efficiency in uncertain environments with large state spaces.

Ax Kabir Kang, Stephen Mussmann 5/6/2026

Instance-Level Costs for Nuanced Classifier Evaluation

arXiv paper proposing Normalized Excess Cost (NEC) metric for classification evaluation that weights errors by per-example costs for safety-critical applications.

Ax Adrian Grassi 5/6/2026

OCRR: A Benchmark for Online Correction Recovery under Distribution Shift

Benchmark for measuring classifier recovery speed under distribution shift and online corrections, addressing real-world deployment scenarios.

Ax Michael A. Riegler, Birk Sebastian Frostelid Torpmann-Hagen 5/6/2026

Pairwise matrices for sparse autoencoders: single-feature inspection mislabels causal axes

Pairwise matrix protocol for sparse autoencoder interpretability reveals limitations of standard single-feature inspection on language model feature analysis.

Ax Yash Aggarwal, Atmika Gorti, Vinija Jain, Aman Chadha, Krishnaprasad Thirunarayan, Manas Gaur 5/6/2026

Moral Sensitivity in LLMs: A Tiered Evaluation of Contextual Bias via Behavioral Profiling and Mechanistic Interpretability

Framework for evaluating bias in LLMs through behavioral profiling and mechanistic interpretability, introducing Moral Sensitivity Index metric.

Ax Amirhossein Yavari, Farnaz Zamani Esfahlani 5/6/2026

Beyond Activation Alignment: The Geometry of Neural Sensitivity

Theoretical analysis of activation alignment methods (RSA, CCA, CKA) for comparing neural representations in biological and artificial systems.

Ax Prakhar Gupta, Garv Shah, Donghua Zhang 5/6/2026

Self-Mined Hardness for Safety Fine-Tuning

Self-Mined Hardness method for LLM safety fine-tuning that scores prompt difficulty by model jailbreak frequency, then trains on hardest prompts.

Ax Chen Huang, Xianhang Li, Vimal Thilak, Etai Littwin, Josh Susskind 5/6/2026

Text-Conditional JEPA for Learning Semantically Rich Visual Representations

Text-Conditional JEPA uses image captions to reduce prediction uncertainty in masked visual representation learning.

Ax Seunghyun Ji 5/6/2026

Ortho-Hydra: Orthogonalized Experts for DiT LoRA

Ortho-Hydra addresses style bleed in LoRA fine-tuning of diffusion transformers using orthogonalized mixture-of-experts architecture.

Ax Anna Sokol, Marianna B. Ganapini, Nitesh V. Chawla 5/6/2026

Do LLMs have core beliefs?

Examines whether LLMs exhibit core beliefs—foundational commitments that resist change—parallel to human cognition.

Ax Gabriel Garcia 5/6/2026

The Right Answer, the Wrong Direction: Why Transformers Fail at Counting and How to Fix It

Investigates why transformers fail at counting tasks, finding failures stem from token generation rather than internal representation, with proposed fixes.

Ax Md Raihan Uddin, Tolunay Seyfi, Fatemeh Afghah 5/6/2026

RFPrompt: Prompt-Based Expert Adaptation of the Large Wireless Model for Modulation Classification

RFPrompt enables efficient adaptation of wireless foundation models to out-of-distribution modulation classification tasks via prompt-based expert adaptation.

Ax Jingjing Zhou, Yongshuai Yang, Qing Qing, Ziqi Xu, Xikun Zhang, Renqiang Luo, Ivan Lee, Feng Xia 5/6/2026

Stable Multimodal Graph Unlearning via Feature-Dimension Aware Quantile Selection

Graph unlearning technique using feature-dimension aware quantile selection for privacy-preserving multimodal graph learning.

Ax Nawapon Sangsiri, Yufei Tao 5/6/2026

Distributed Learning with Adversarial Gradient Perturbations

Studies convex optimization under adversarial gradient perturbations in distributed learning, analyzing privacy-utility tradeoffs.

Ax Hongbo Jin, Rongpeng Zhu, Zhongjing Du, Xu Jiang, Jingqi Tian, Qiaoman Zhang, Jiayu Ding 5/6/2026

DGPO: Distribution Guided Policy Optimization for Fine Grained Credit Assignment

DGPO algorithm for fine-grained credit assignment in LLM reinforcement learning, improving reasoning step isolation in chain-of-thought generations.

Ax Ahmadreza Eslaminia, Chuhan Cai, Cameron Smith, Ruo-Syuan Mei, Shichen Li, Rajiv Malhotra, Klara Nahrstedt, Chenhui Shao 5/6/2026

LLM-ADAM: A Generalizable LLM Agent Framework for Pre-Print Anomaly Detection in Additive Manufacturing

LLM-ADAM agent framework for detecting anomalies in additive manufacturing pre-prints, helping users without manufacturing expertise catch process-planning errors.

Ax Hahyeon Choi, Nojun Kwak 5/6/2026

Toward Structural Multimodal Representations: Specialization, Selection, and Sparsification via Mixture-of-Experts

S3 framework decomposes multimodal inputs into semantic experts with selective routing and sparsification via mixture-of-experts approach.

Ax Gr\'egoire Lambrecht, Mathieu Lauri\`ere 5/6/2026

Population-Aware Imitation Learning in Mean-field Games with Common Noise

Addresses imitation learning in mean-field games with stochastic population distribution, developing algorithms for population-aware agent policies.

Ax Zhe Li, Bicheng Ying, Zidong Liu, Haibo Yang 5/6/2026

Learning Dynamics of Zeroth-Order Optimization: A Kernel Perspective

Investigates zeroth-order optimization methods for fine-tuning large language models, explaining why they work despite theoretical dimension-dependent slowdown predictions.

Ax Rishi Raj Sahoo, Subhankar Mishra 5/6/2026

GRAFT: Auditing Graph Neural Networks via Global Feature Attribution

Proposes GRAFT, a global explanation framework for Graph Neural Networks that identifies which input node attributes drive model predictions.

Ax Yi Liu 5/6/2026

Two Calls, Two Moments, and the Vote-Accuracy Curve of Repeated LLM Inference

Studies how repeated LLM inference sampling improves accuracy beyond single-call performance, analyzing correctness correlation across examples through latent probability distributions.

Ax Rishi Raj Sahoo, Jyotirmaya Shivottam, Subhankar Mishra 5/6/2026

Graph Reconstruction from Differentially Private GNN Explanations

Research shows differential privacy-perturbed GNN explanations can leak graph structure through adversarial reconstruction attacks.

Ax Akshat Singh Jaswal, Ashish Baghel, Paras Chopra 5/6/2026

Discovering Reinforcement Learning Interfaces with Large Language Models

Uses LLMs to automatically synthesize complete RL task interfaces including observations and rewards from natural language.

Ax Doojin Baek, Gyubin Lee, Junyeob Baek, Hosung Lee, Sungjin Ahn 5/6/2026

Learning to Theorize the World from Observation

Proposes learning-to-theorize approach where agents build internal theories from observations inspired by cognitive science.

Ax Duc Dm, Thao Do, Minh Son Hoang, Anh Le Duc Tran, Daeyoung Kim, Huy Nguyen 5/6/2026

FIBER: A Differentially Private Optimizer with Filter-Aware Innovation Bias Correction

Introduces FIBER optimizer correcting bias from filtering in differentially private training with adaptive optimizers.

Ax Al Zadid Sultan Bin Habib, Gianfranco Doretto, Donald A. Adjeroh 5/6/2026

DynaTab: Dynamic Feature Ordering as Neural Rewiring for High-Dimensional Tabular Data

Proposes DynaTab, architecture with dynamic feature ordering for high-dimensional tabular data using neural rewiring.

Ax Yu-Ting Lee, Samuel Yen-Chi Chen, Fu-Chieh Chang 5/6/2026

Quantum Hierarchical Reinforcement Learning via Variational Quantum Circuits

Studies hierarchical reinforcement learning using variational quantum circuits.

Ax Zhenghao Huang, Huilin Yao, Kaikai Wang, Lin Shu 5/6/2026

Learning Generalizable Action Representations via Pre-training AEMG

Proposes AEMG, first large-scale self-supervised framework for learning generalizable EMG representations across subjects and devices.

Ax Michal Valko 5/6/2026

Adaptive graph-based algorithms for conditional anomaly detection and semi-supervised learning

Develops fast approximate online graph-based algorithms for semi-supervised learning and anomaly detection.

Ax James Rowbottom, Elizabeth L. Baker, Nick Huang, Ben Adcock, Carola-Bibiane Sch\"onlieb, Alexander Denker 5/6/2026

GRIFDIR: Graph Resolution-Invariant FEM Diffusion Models in Function Spaces over Irregular Domains

Proposes resolution-invariant diffusion models for function spaces over irregular domains using score-based approach.

Ax Zhao Wei, Kenneth Hor Cheng Koh, Sheng Yuan Chin, James Chun Yip Chan, Chin Chun Ooi, Yew-Soon Ong 5/6/2026

Meta-Inverse Physics-Informed Neural Networks for High-Dimensional Ordinary Differential Equations

Applies physics-informed neural networks with meta-learning to solve inverse problems in high-dimensional ODEs.