Ax Oussama Bouanani, Jim Berend, Wojciech Samek, Sebastian Lapuschkin, Maximilian Dreyer 4/27/2026

Contrastive Semantic Projection: Faithful Neuron Labeling with Contrastive Examples

Research on neuron labeling for interpretability using contrastive examples to produce faithful textual descriptions of deep network internal units.

Ax Thibault Pautrel, Florent Bouchard, Ammar Mian, Guillaume Ginolhac 4/27/2026

FedSPDnet: Geometry-Aware Federated Deep Learning with SPDnet

Research paper on federated learning frameworks for SPDnet models operating on symmetric positive definite matrices with geometry-preserving aggregation strategies.

Ax Coenraad Mouton, Randle Rabe, Niklas C. Koser, Nicolai Krekiehn, Christopher Hansen, Jan-Bernd H\"ovener, Claus-C. Gl\"uer 4/27/2026

Useful nonrobust features are ubiquitous in biomedical images

Studies nonrobust predictive features in deep medical imaging models. Shows networks learn adversarially vulnerable patterns useful in-distribution.

Ax Jeffrey N\"af, Riana Valera Mbelson, Markus Meierer 4/27/2026

CLVAE: A Variational Autoencoder for Long-Term Customer Revenue Forecasting

Variational autoencoder for long-term customer revenue forecasting from sparse transaction data. Combines probabilistic and ML approaches.

Ax Rajinder Sandhu, Di Mu, Cheng Chang, Md Shahriar Tasjid, Himanshu Rai, Maksims Volkovs, Ga Wu 4/27/2026

Aligning Dense Retrievers with LLM Utility via DistillationAligning Dense Retrievers with LLM Utility via Distillation

Framework aligning dense retrievers with LLM utility via distillation for improved RAG performance. Balances precision and computational cost.

Ax Calvin Tsay 4/27/2026

Relaxation-Informed Training of Neural Network Surrogate Models

Training method for neural network surrogate models that improves their embedding in optimization problems. Enhances tractability of MILP formulations.

Ax Marlon Tobaben, Talal Alrawajfeh, Marcus Klasson, Mikko Heikkil\"a, Arno Solin, Antti Honkela 4/27/2026

Privacy Leakage via Output Label Space and Differentially Private Continual Learning

Studies privacy side-channels in ML models via output label space and proposes differentially private continual learning defenses.

Ax Anoop Cherian, Radu Corcodel, Siddarth Jain, Diego Romeres 4/27/2026

LLMPhy: Parameter-Identifiable Physical Reasoning Combining Large Language Models and Physics Engines

LLMPhy framework combines LLMs with physics simulators for physical reasoning and parameter identification in robotic and collision avoidance tasks.

Ax Akansha Kalra, Basavasagar Patil, Guanhong Tao, Daniel S. Brown 4/27/2026

How Vulnerable Is My Learned Policy? Universal Adversarial Perturbation Attacks On Modern Behavior Cloning Policies

Systematic study of adversarial attacks on behavior cloning policies including BC, LSTM-GMM, IBC, and diffusion-based imitation learning.

Ax Illia Horenko, Davide Bassetti, Luk\'a\v{s} Posp\'i\v{s}il 4/27/2026

Fast, close, non-singular and property-preserving approximations of entropic measures

Fast approximations of entropic measures like Shannon entropy and KL divergence for physics, information theory, and ML applications.

Ax Zehua Pei, Ying Zhang, Hui-Ling Zhen, Tao Yuan, Xianzhi Yu, Zhenhua Dong, Sinno Jialin Pan, Mingxuan Yuan, Bei Yu 4/27/2026

PreMoE: Proactive Inference for Efficient Mixture-of-Experts

PreMoE: training-free framework to compile sparse Mixture-of-Experts variants for deployment-specific optimization using predicted expert utility.

Ax Aideen Fay, In\'es Garc\'ia-Redondo, Qiquan Wang, Haim Dubossarsky, Anthea Monod 4/27/2026

The Shape of Adversarial Influence: Characterizing LLM Latent Spaces with Persistent Homology

Persistent homology characterizes high-dimensional geometry and topology of LLM representation spaces under adversarial inputs.

Ax Yijia Dai, Zhaolin Gao, Yahya Sattar, Sarah Dean, Jennifer J. Sun 4/27/2026

Pre-trained Large Language Models Learn Hidden Markov Models In-context

Pre-trained LLMs learn to model Hidden Markov Model behavior through in-context learning without fine-tuning.

Ax Omkar Tupe, Max Hartman, Lav R. Varshney, Saurav Prakash 4/27/2026

Federated Nonlinear System Identification

Federated learning framework for nonlinear system identification with theoretical convergence guarantees improving with client count.

Ax Peng Chen, Jiaji Zhang, Hailiang Zhao, Yirong Zhang, Shenyao Chen, Jiahong Yu, Xueyan Tang, Yixuan Wang, Hao Li, Jianping Zou, Gang Xiong, Kingsum Chow, Shuibing He, Shuiguang Deng 4/27/2026

Toward Robust and Efficient ML-Based GPU Caching for Modern Inference

Learning-augmented caching algorithm for GPU inference that combines ML predictors with robustness guarantees against prediction errors.

Ax Shirin Alanova, Kristina Kazistova, Ekaterina Galaeva, Alina Kostromina, Vladimir Smirnov, Redko Dmitry, Alexey Dontsov, Maxim Zhelnin, Evgeny Burnaev, Egor Shvetsov 4/27/2026

Motivating Next-Gen Accelerators with Flexible (N:M) Activation Sparsity via Benchmarking Lightweight Post-Training Sparsification Approaches

Comprehensive analysis of post-training N:M activation sparsification methods for efficient LLM inference and dynamic compression.

Ax Mingxing Rao, Bowen Qu, Daniel Moyer 4/27/2026

Score-based Membership Inference on Diffusion Models

Score-based membership inference attacks on diffusion models using predicted noise vectors with reduced computational overhead.

Ax Bruno Puri, Jim Berend, Sebastian Lapuschkin, Wojciech Samek 4/27/2026

Atlas-Alignment: Making Interpretability Transferable Across Language Models

Atlas-Alignment enables interpretability methods to transfer across different LLMs, reducing cost of model-specific interpretability pipelines.

Ax Reginald Zhiyan Chen, Heng-Sheng Chang, Prashant G. Mehta 4/27/2026

Differentiable Filtering for Learning Hidden Markov Models

Belief Net: differentiable filtering framework for learning Hidden Markov Model parameters from sequential data.

Ax Promise Ekpo, Saesha Agarwal, Felix Grimm, Lekan Molu, Angelique Taylor 4/27/2026

AdaFair-MARL: Enforcing Adaptive Fairness Constraints in Multi-Agent Reinforcement Learning

AdaFair-MARL enforces adaptive fairness constraints in multi-agent reinforcement learning systems without fixed penalties.

Ax Kairong Luo, Zhenbo Sun, Haodong Wen, Xinyu Shi, Jiarui Cui, Chenyi Dang, Kaifeng Lyu, Wenguang Chen 4/27/2026

How Learning Rate Decay Wastes Your Best Data in Curriculum-Based LLM Pretraining

Study on how learning rate decay interacts with curriculum-based pretraining in LLMs, showing data quality utilization inefficiencies.

Ax Henrijs Princis, Arindam Sharma, Cristina David 4/27/2026

TreeCoder: Systematic Exploration and Optimisation of Decoding and Constraints for LLM Code Generation

TreeCoder framework for exploring decoding strategies and constraints in LLM code generation to enforce correctness during inference rather than post-hoc.

Ax Rebonto Haque, Oliver M. Turnbull, Anisha Parsan, Nithin Parsan, John J. Yang, Anna L. Beukenhorst, Charlotte M. Deane 4/27/2026

Mechanistic Interpretability of Antibody Language Models Using SAEs

Mechanistic interpretability of antibody language models using sparse autoencoders for feature discovery and steering in protein sequence generation.

Ax Marcel Meyer, Sascha Kaltenpoth, Henrik Albers, Kevin Zalipski, Oliver M\"uller 4/27/2026

TS-Arena -- A Live Forecast Pre-Registration Platform

TS-Arena live forecasting platform for evaluating time series foundation models on unknown future data, addressing train-test contamination issues.

Ax Xiucheng Xu, Bingbing Xu, Xueyun Tian, Zihe Huang, Rongxin Chen, Yunfan Li, Huawei Shen 4/27/2026

Chain-of-Memory: Lightweight Memory Construction with Dynamic Evolution for LLM Agents

Chain-of-Memory proposes lightweight dynamic external memory for LLM agents with efficient construction and adaptive retrieval-augmented generation.

Ax Deming Chen, Vijay Ganesh, Weikai Li, Yingyan Celine Lin, Yong Liu, Subhasish Mitra, David Z. Pan, Ruchir Puri, Jason Cong, Yizhou Sun 4/27/2026

Report for NSF Workshop on AI for Electronic Design Automation

NSF workshop report on AI for Electronic Design Automation discussing LLMs, GNNs, RL, and neurosymbolic methods for EDA automation.

Ax Canyang Zhao, Bolin Peng, J. Patrick Mayo, Ce Ju, Bing Liu 4/27/2026

Cross-Session Decoding of Neural Spiking Data via Task-Conditioned Latent Alignment

Task-conditioned latent alignment framework for cross-session neural decoding with limited data via autoencoder-based transfer learning.

Ax Amin Oji, Paul Fieguth 4/27/2026

Joint Embedding Variational Bayes

Variational Joint Embedding framework for non-contrastive self-supervised learning using symmetric conditional ELBO on paired encoder embeddings.

Ax Noor Islam S. Mohammad, Md Muntaqim Meherab 4/27/2026

Regularized Meta-Learning for Improved Generalization

Regularized meta-learning framework addressing redundancy, multicollinearity, and overfitting in deep ensemble methods through four-stage projection pipeline.

Ax Ofir Gordon, Lior Dikstein, Arnon Netzer, Idan Achituve, Hai Victor Habi 4/27/2026

LATMiX: Learnable Affine Transformations for Microscaling Quantization of LLMs

LATMiX introduces learnable affine transformations for post-training quantization of LLMs, extending beyond rotation/Hadamard-based approaches to reduce activation outliers.

Ax Aleena Siji, Amir Mohammad Karimi Mamaghan, Ferdinand Kapl, Tobias H\"oppe, Emmanouil Angelis, Andrea Dittadi, Maurice Brenner, Michael Heinzinger, Karl Henrik Johansson, Kaitlin Maile, Johannes von Oswald, Stefan Bauer 4/27/2026

From Words to Amino Acids: Does the Curse of Depth Persist?

Investigates curse of depth in protein language models, comparing depth scaling effects between protein and natural language transformers.

Ax Junyi An, Chao Qu, Yun-Fei Shi, Zhijian Zhou, Fenglei Cao, Yuan Qi 4/27/2026

Equivariant Asynchronous Diffusion: An Adaptive Denoising Schedule for Accelerated Molecular Conformation Generation

Equivariant asynchronous diffusion method for adaptive denoising schedules in 3D molecular structure generation balancing sequential and hierarchical constraints.

Ax Md Muntaqim Meherab, Noor Islam S. Mohammad, Faiza Feroz 4/27/2026

Causal Concept Graphs in LLM Latent Space for Stepwise Reasoning

Causal Concept Graphs combine sparse autoencoders with differentiable structure learning to capture causal dependencies between interpretable latent features for multi-step LLM reasoning.

Ax Luca Ambrogioni 4/27/2026

How Out-of-Equilibrium Phase Transitions can Seed Pattern Formation in Trained Diffusion Models

Theoretical framework explaining pattern formation in diffusion models through out-of-equilibrium phase transitions and denoising dynamics instabilities.

Ax Rajveer Singh 4/27/2026

SpectralLoRA: Is Low-Frequency Structure Sufficient for LoRA Adaptation? A Spectral Analysis of Weight Updates

SpectralLoRA analyzes spectral structure of LoRA weight updates via DCT, showing low-frequency components dominate adaptation across BERT and RoBERTa models.

Ax Hongtao Xu, Jianchao Tan, Yuxuan Hu, Pengju Lu, Hongyu Wang, Pingwei Sun, Yerui Sun, Yuchen Xie, Xunliang Cai, Mingzhen Li, Weile Jia 4/27/2026

SparseBalance: Load-Balanced Long Context Training with Dynamic Sparse Attention

SparseBalance addresses load balancing challenges in distributed sparse attention training for long-context LLMs by co-optimizing sequence length heterogeneity and sparsity sensitivity.

Ax Anurita Das 4/27/2026

MCAP: Deployment-Time Layer Profiling for Memory-Constrained LLM Inference

MCAP load-time profiling method for memory-constrained LLM inference enabling dynamic precision and tier selection across heterogeneous hardware.

Ax Zhaokun Wang, Jinyu Guo, Jingwen Pu, Hongli Pu, Meng Yang, Xunlei Chen, Jie Ou, Wenyi Li, Guangchun Luo, Wenhong Tian 4/27/2026

CAP: Controllable Alignment Prompting for Unlearning in LLMs

CAP method enables selective knowledge unlearning in closed-source LLMs through controllable alignment prompting without model weight access.

Ax Noel Elias, Homa Esfahanizadeh, Kaan Kale, Sriram Vishwanath, Muriel Medard 4/27/2026

MultiTok: Variable-Length Tokenization for Efficient LLMs Adapted from LZW Compression

MultiTok tokenization method inspired by LZW compression for variable-length encoding to reduce LLM training resource requirements.

Ax Haiyu Deng, Yanna Jiang, Guangsheng Yu, Qin Wang, Xu Wang, Baihe Ma, Wei Ni, Ren Ping Liu 4/27/2026

PoLO: Proof-of-Learning and Proof-of-Ownership at Once with Chained Watermarking

PoLO protocol combines proof-of-learning and proof-of-ownership using chained watermarking with 99% detection accuracy and reduced verification costs.

Ax C\'eline Finet, Stephane Da Silva Martins, Jean-Bernard Hayet, Ioannis Karamouzas, Javad Amirian, Sylvie Le H\'egarat-Mascle, Julien Pettr\'e, Emanuel Aldea 4/27/2026

Recent Advances in Multi-Agent Human Trajectory Prediction: A Comprehensive Review

Survey of deep learning methods for multi-agent human trajectory prediction with applications in robotics, autonomous driving, and crowd modeling.

Ax Shuyi Lin, Anshuman Suri, Alina Oprea, Cheng Tan 4/27/2026

Toward Principled LLM Safety Testing: Solving the Jailbreak Oracle Problem

Formalizes jailbreak oracle problem and introduces principled methods for systematic LLM safety testing and vulnerability assessment.

Ax Do-hyeon Yoon, Minsoo Chun, Thomas Allen, Hans M\"uller, Min Wang, Rajesh Sharma 4/27/2026

Intrinsic Fingerprint of LLMs: Continue Training is NOT All You Need to Steal A Model!

Proposes robust LLM fingerprinting approach for copyright protection that resists continued training and model theft attacks.

Ax Xingyu Shen, Yingfa Chen, Zhen Leng Thai, Xu Han, Zhiyuan Liu, Maosong Sun 4/27/2026

StateX: Enhancing RNN Recall via Post-training State Expansion

StateX method enhances RNN recall for long-context tasks by expanding fixed-size recurrent state through post-training state expansion.

Ax Yinglun Zhu, Jiancheng Zhang, Fuzhi Tang 4/27/2026

Test-Time Matching: Unlocking Compositional Reasoning in Multimodal Models

Addresses compositional reasoning limitations in multimodal AI models by introducing improved evaluation metrics and group matching scoring.

Ax Rickmer Krohn, Vignesh Prasad, Gabriele Tiboni, Georgia Chalvatzaki 4/27/2026

Self-Supervised Multisensory Pretraining for Contact-Rich Robot Reinforcement Learning

MSDP framework for self-supervised multisensory representation learning combining vision, force, and proprioception for contact-rich robot manipulation.

Ax Sajad Movahedi, Timur Carstensen, Arshia Afzal, Frank Hutter, Antonio Orvieto, Volkan Cevher 4/27/2026

Selective Rotary Position Embedding

Selective RoPE: position encoding combining fixed-angle rotations from RoPE with input-dependent selective gating for improved language modeling.

Ax Rashmeet Kaur Nayyar, Naman Shah, Siddharth Srivastava 4/27/2026

Context-Sensitive Abstractions for Reinforcement Learning with Parameterized Actions

RL approach for parameterized action spaces combining discrete action selection with continuous parameter learning without hand-crafted action models.

Ax Sourav Saha, Mandar Mitra, Aditya Dutta 4/27/2026

LLMs as Assessors: Right for the Right Reason?

Study of whether LLMs as relevance assessors in IR tasks provide correct reasoning, extending research on LLMs as judges for output evaluation.

Ax John Kirchenbauer, Abhimanyu Hans, Brian Bartoldson, Micah Goldblum, Ashwinee Panda, Tom Goldstein 4/27/2026

Multi-Token Prediction via Self-Distillation

Self-distillation approach to convert pretrained autoregressive language models to multi-token prediction for faster inference without auxiliary models.