Ax Oubo Ma, Ruixiao Lin, Yang Dai, Jiahao Chen, Chunyi Zhou, Linkang Du, Shouling Ji 5/15/2026

Angel or Demon: Investigating the Plasticity Interventions' Impact on Backdoor Threats in Deep Reinforcement Learning

Studies impact of plasticity interventions on backdoor attack vulnerabilities in deep reinforcement learning agents.

Ax Zhipeng Zhang 5/15/2026

Silent Collapse in Recursive Learning Systems

Identifies silent collapse phenomenon in recursive learning systems where models trained on self-generated data degrade undetected by standard metrics.

Ax Andreas Schlaginhaufen, Maryam Kamgarpour 5/15/2026

Fast Rates for Inverse Reinforcement Learning

Statistical results for entropy-regularized inverse reinforcement learning with linear reward classes in finite-horizon MDPs.

Ax Sohaib Afifi 5/15/2026

An Amortized Efficiency Threshold for Comparing Neural and Heuristic Solvers in Combinatorial Optimization

Analysis of energy efficiency comparing neural combinatorial optimization solvers to CPU metaheuristics accounting for training costs.

Ax Manrui Jiang, Jingru Huang, Yong Chen, Chen Zhang 5/15/2026

DRL-STAF: A Deep Reinforcement Learning Framework for State-Aware Forecasting of Complex Multivariate Hidden Markov Processes

Deep reinforcement learning framework combining HMMs with neural networks for forecasting multivariate hidden Markov processes.

Ax Shin So, Kyelim Lee, Albert No 5/15/2026

Slower Generalization, Faster Memorization: A Sweet Spot in Algorithmic Learning

Study of grokking phenomenon showing Transformers achieve fastest validation accuracy at intermediate dataset sizes, not largest ones.

Ax Nabil Iqbal, T. Anderson Keller, Yue Song, Takeru Miyato, Max Welling 5/15/2026

Spontaneous symmetry breaking and Goldstone modes for deep information propagation

Studies Goldstone-like modes in equivariant deep neural networks, demonstrating how symmetry breaking enables coherent information propagation across layers.

Ax Davide Scassola, Andrea Coser, Sebastiano Saccani 5/15/2026

ReMIA: a Powerful and Efficient Alternative to Membership Inference Attacks against Synthetic Data Generators

Proposes ReMIA, efficient membership inference attack against synthetic data generators that avoids shadow model overhead for privacy evaluation.

Ax Tommaso Mencattini, Francesco Montagna, Francesco Locatello 5/15/2026

The Rate-Distortion-Polysemanticity Tradeoff in SAEs

Characterizes tradeoff between reconstruction accuracy, feature efficiency, and interpretability in sparse autoencoders for mechanistic interpretability.

Ax Konstantinos Kontras, Trui Osselaer, Stylianos G. Mouslech, Angeliki-Ilektra Karaiskou, Guido Gagliardi, Thomas Strypsteen, Mohammad Hossein Badiei, Anku Rani, Maarten Vanmarcke, Miguel Bhagubai, Chanakya Ekbote, Jaedong Hwang, Christos Chatzichristos, Paul Pu Liang, Maarten De Vos 5/15/2026

NeuroAtlas: Benchmarking Foundation Models for Clinical EEG and Brain-Computer Interfaces

Benchmarks foundation models on clinical EEG tasks and brain-computer interfaces with standardized evaluation protocols addressing dataset and preprocessing variations.

Ax Krish Sharma, Omar Naim, Soumadeep Saha, Nicholas Asher 5/15/2026

TAPIOCA: Why Task- Aware Pruning Improves OOD model Capability

Analyzes task-aware layer pruning effects on model generalization, showing improvements in out-of-distribution accuracy across regression tasks and LLMs.

Ax Bat-Sheva Einbinder, Hen Davidov, Yee Whye Teh, Yarin Gal, Yaniv Romano 5/15/2026

Selective Safety Steering via Value-Filtered Decoding

Proposes value-filtered decoding method to improve LLM safety by selectively steering generation away from unsafe outputs at inference time.

Ax Qirui Liu, Hao Chen, Weijie Shi, Jiajie Xu, Jia Zhu 5/15/2026

Cognitive-Uncertainty Guided Knowledge Distillation for Accurate Classification of Student Misconceptions

Uses knowledge distillation with cognitive-uncertainty guidance to improve classification of student misconceptions in educational settings.

Ax Hongyu Lin, Antonio Briola, Yuanrong Wang, Tomaso Aste 5/15/2026

Compositional Sparsity as an Inductive Bias for Neural Architecture Design

Investigates compositional sparsity as structural prior enabling deep neural networks to overcome curse of dimensionality using Information Filtering Networks.

Ax Suorong Yang, Hanqi Zhu, Hai Gan, Fangjian Su, Guang Li, Furao Shen, Soujanya Poria 5/15/2026

Beyond What to Select: A Plug-and-play Oscillatory Data-Volume Scheduling for Efficient Model Training

Proposes oscillatory data-volume scheduling for efficient model training by dynamically adjusting both sample selection and data volume throughout training.

Ax Byeongchan Kim, Min-hwan Oh 5/15/2026

Peng's Q($\lambda$) for Conservative Value Estimation in Offline Reinforcement Learning

Model-free offline RL algorithm using Peng's Q-lambda operator for conservative multi-step value estimation without model learning.

Ax Alberto Tamajo, Srinandan Dasmahapatra, Rahman Attar 5/15/2026

Understanding Imbalanced Forgetting in Rehearsal-Based Class-Incremental Learning

Analysis of imbalanced forgetting in class-incremental learning explaining why balanced rehearsal still causes differential class forgetting.

Ax Yan Jiang, Ruihong Qiu, Zi Huang 5/15/2026

GFMate: Empowering Graph Foundation Models with Test-time Prompt Tuning

Test-time prompt tuning framework for graph foundation models enabling cross-domain adaptation with trainable auxiliary prompts.

Ax Sohom Mukherjee, Anh-Duy Pham, Richard Pibernik, Yunbei Xu 5/15/2026

In-Context Learning for Data-Driven Censored Inventory Control

In-context learning approach for censored inventory control using LLM predictions in decision-dependent censoring settings.

Ax Paolo Mandica, Micha{\l} Brzozowski, Zuzanna Dubanowska, Neo Christopher Chung 5/15/2026

GPart: End-to-End Isometric Fine-Tuning via Global Parameter Partitioning

Parameter-efficient fine-tuning method improving upon LoRA via isometric global parameter partitioning preserving optimization landscape.

Ax Thomas Witt 5/15/2026

XFP: Quality-Targeted Adaptive Codebook Quantization with Sparse Outlier Separation for LLM Inference

Dynamic weight quantization for LLM inference with quality-targeted adaptive codebooks and sparse outlier separation for efficient deployment.

Ax Kamil Ciosek, Aleksandr V. Petrov, Nicol\`o Felicioni, Konstantina Palla 5/15/2026

Fast Adversarial Attacks with Gradient Prediction

Fast adversarial attack method eliminating backward pass cost by predicting input gradients from forward hidden states via linear regression.

Ax Magdalena Proszewska, N. Siddharth 5/15/2026

AIMing for Standardised Explainability Evaluation in GNNs: A Framework and Case Study on Graph Kernel Networks

Framework for standardized explainability evaluation in graph neural networks with emphasis on inherently interpretable models.

Ax Senne Deproost, Denis Steckelmacher, Ann Now\'e 5/15/2026

Critic-Driven Voronoi-Quantization for Distilling Deep RL Policies to Explainable Models

Policy distillation method using Voronoi quantization and critic guidance to convert deep RL policies into interpretable surrogate models.

Ax Leonardo Ferreira Guilhoto, Akshat Kaushal, Paris Perdikaris 5/15/2026

A Mutual Information Lower Bound for Multimodal Regression Active Learning

Active learning acquisition function using mutual information framework for multimodal regression with epistemic uncertainty separation.

Ax Earl Killian 5/15/2026

A Hardware-Aware, Per-Layer Methodology for Post-Training Quantization of Large Language Models

Post-training quantization methodology for LLMs using Scaled Outer Product achieving near-lossless 4.5-6 bit compression with hardware awareness.

Ax Yuehao Liu, Shanyan Guan, Weijia Zhang, Xuanming Shang, Yanhao Ge, Wei Li, Chao Ma 5/15/2026

Octopus: History-Free Gradient Orthogonalization for Continual Learning in Multimodal Large Language Models

Continual learning method for multimodal LLMs using gradient orthogonalization to mitigate catastrophic forgetting without storing historical data.

Ax Sreenivas Gollapudi, Kostas Kollias, Kamesh Munagala, Ali Sinop 5/15/2026

Efficient Online Conformal Selection with Limited Feedback

Online conformal selection algorithm for efficiently selecting minimal option subsets under limited feedback with pre-specified success probability guarantees.

Ax Mahdi Sabbaghi, George Pappas, Adel Javanmard, Hamed Hassani 5/15/2026

InfoSFT: Learn More and Forget Less with Information-Aware Token Weighting

InfoSFT method for supervised fine-tuning of LLMs using information-aware token weighting to reduce overfitting and minimize policy shift from base model.

Ax Ao Xu, Tieru Wu 5/15/2026

Distance-Matrix Wasserstein Statistics for Scalable Gromov--Wasserstein Learning

Distance-Matrix Wasserstein (DMW) statistics for scalable Gromov-Wasserstein learning on graphs, shapes, and point clouds without common coordinate systems.

Ax Sanjeev Manivannan, Shuban V 5/15/2026

Second-Order Actor-Critic Methods for Discounted MDPs via Policy Hessian Decomposition

Second-order actor-critic methods for reinforcement learning via policy Hessian decomposition, providing curvature-aware updates for faster convergence.

Ax Kai Yan, Alexander G. Schwing, Yu-Xiong Wang 5/15/2026

Boosting Reinforcement Learning with Verifiable Rewards via Randomly Selected Few-Shot Guidance

Method for improving sample efficiency in reinforcement learning from verifiable rewards by using randomly selected few-shot guidance for LLM chain-of-thought tasks.

Ax Kiljae Lee, Ziqi Liu, Weijing Tang, Yuan Zhang 5/15/2026

Generalized Priority-Aware Shapley Value

Generalized priority-aware Shapley value (GPASV) extending Shapley value methods to arbitrary directed weighted priority graphs for data valuation.

Ax Zara Zetlin, Kayhan Moharreri, Maria Safi 5/15/2026

TopoPrimer: The Missing Topological Context in Forecasting Models

TopoPrimer framework incorporating global topological structure via persistent homology into time series forecasting models, improving accuracy and handling cold-start problems.

Ax Linghao Kong, Megan Flynn, Michael Peng, Nir Shavit, Mark Kurtz, Alexandre Marques 5/15/2026

An Interpretable Latency Model for Speculative Decoding in LLM Serving

Interpretable latency model for speculative decoding in LLM serving systems, analyzing performance under varying request loads and dynamic batch sizes.

Ax Anurup Ganguli 5/15/2026

TFGN: Task-Free, Replay-Free Continual Pre-Training Without Catastrophic Forgetting at LLM Scale

TFGN architecture for continual pre-training of LLMs on heterogeneous text domains without replay buffers or task labels, solving catastrophic forgetting at scale.

Ax Quanhao Li, Junqiu Yu, Kaixun Jiang, Yujie Wei, Zhen Xing, Pandeng Li, Ruihang Chu, Shiwei Zhang, Yu Liu, Zuxuan Wu 5/15/2026

DiffusionOPD: A Unified Perspective of On-Policy Distillation in Diffusion Models

DiffusionOPD paradigm for multi-task reinforcement learning on diffusion-based text-to-image models, addressing cross-task interference and catastrophic forgetting.

Ax Rafi Al Attrach, Rajna Fani, Sebastian Lobentanzer, Joan Giner-Miguelez, Debanshu Das, Varuni H. K., Nobin Sarwar, Rajat Ghosh, Anwai Archit, Surbhi Motghare, Christina Conrad Parry, Luis Oala, Lara Grosso, Joaquin Vanschoren, Steffen Vogler, Sujata Goswami, Eric S. Rosenthal, Marzyeh Ghassemi, Matthew McDermott, Tom Pollard 5/15/2026

Croissant Baker: Metadata Generation for Discoverable, Governable, and Reusable ML Datasets

Croissant Baker tool for automated metadata generation of ML datasets using JSON-LD format, improving dataset discoverability and reproducibility across platforms.

Ax Daniel Asare Kyei, Alimatu Saadia-Yussiff, Maame G. Asante-Mensah, Abdul Lateef-Yussiff, Charles Roland Haruna, Derry Emmanuel 5/15/2026

Novel Dynamic Batch-Sensitive Adam Optimiser for Vehicular Accident Injury Severity Prediction

Dynamic Batch-Sensitive Adam optimizer for handling imbalanced and sequential datasets in vehicular accident injury severity prediction tasks.

Ax Yang Li, Erik Nijkamp, Semih Yavuz, Shafiq Rayhan Joty 5/15/2026

Learning from Language Feedback via Variational Policy Distillation

Variational policy distillation method for reinforcement learning from language feedback, improving sample efficiency on complex reasoning tasks with dense token-level supervision.

Ax Frederik Schmitt, Matthias Cosler, Niklas Metzger, Julian Siber, Vladimir Krsmanovic, Mohamed Ghanem, Bernd Finkbeiner 5/15/2026

Natural Synthesis: Outperforming Reactive Synthesis Tools with Large Reasoning Models

Neuro-symbolic approach using large reasoning models to solve reactive synthesis for hardware circuit generation, outperforming traditional formal verification tools.

Ax Christopher Stith, Medha Barath, Vahid Balazadeh, Jesse C. Cresswell, Rahul G. Krishnan 5/15/2026

Causal Foundation Models with Continuous Treatments

Foundation models for causal inference with continuous treatments, extending causal estimation beyond binary treatment settings to continuous intervention ranges.

Ax Will Schwarzer, Scott Niekum 5/15/2026

Training ML Models with Predictable Failures

Statistical method for predicting ML model failure rates at deployment scale by extrapolating from largest k failure scores in evaluation sets.

Ax Saisab Sadhu, Pratinav Seth, Vinay Kumar Sankarapu 5/15/2026

Forgetting That Sticks: Quantization-Permanent Unlearning via Circuit Attribution

Research on machine unlearning in quantized LLMs showing that 4-bit post-training quantization can reverse unlearning effects, revealing systematic failures in gradient-based forgetting methods.

Ax Xiaohua Zhan, Kazuki Egashira, Robin Staab, Mark Vero, Martin Vechev 5/15/2026

Widening the Gap: Exploiting LLM Quantization via Outlier Injection

Adversarial attack exploiting LLM quantization via outlier injection, demonstrating security risks in quantized model deployment.

Ax Zhengxi Lu, Zhiyuan Yao, Zhuowen Han, Zi-Han Wang, Jinyang Wu, Qi Gu, Xunliang Cai, Weiming Lu, Jun Xiao, Yueting Zhuang, Yongliang Shen 5/15/2026

Self-Distilled Agentic Reinforcement Learning

On-policy self-distillation method providing dense token-level guidance for multi-turn LLM agent reinforcement learning.

Ax Ellwil Sharma, Arastu Sharma 5/15/2026

Eradicating Negative Transfer in Multi-Physics Foundation Models via Sparse Mixture-of-Experts Routing

Sparse mixture-of-experts routing to mitigate negative transfer in multi-physics foundation models trained on diverse PDE regimes.

Ax Shashwat Goel, Nikhil Chandak, Arvindh Arun, Ameya Prabhu, Steffen Staab, Moritz Hardt, Maksym Andriushchenko, Jonas Geiping 5/15/2026

FutureSim: Replaying World Events to Evaluate Adaptive Agents

Evaluation framework for adaptive AI agents by replaying chronological world events, enabling assessment of real-time adaptation capabilities.

Ax Alexandre Le Mercier, Chris Develder, Thomas Demeester 5/15/2026

Hidden State Poisoning Attacks against Mamba-based Language Models

Study of hidden state poisoning attacks against Mamba state space models, demonstrating amnesia effects from adversarial input phrases.

Ax Eslam Reda, Sara El-Metwally 5/15/2026

A Hormone-inspired Emotion Layer for Transformer language models (HELT)

Emotion layer module for Transformer language models to process emotional context beyond sentiment analysis.