Ax Haolong Chen, Liang Zhang, Zhengyuan Xin, Guangxu Zhu 5/21/2026

STM3: Mixture of Multiscale Mamba for Long-Term Spatio-Temporal Time-Series Prediction

Proposes multiscale Mamba mixture architecture for long-term spatio-temporal time-series prediction with efficient dependency learning.

Ax Geon Lee, Bhuvesh Kumar, Clark Mingxuan Ju, Tong Zhao, Kijung Shin, Neil Shah, Liam Collins 5/21/2026

Sequential Data Augmentation for Generative Recommendation

Studies data augmentation techniques for generative recommendation systems that predict user interactions from historical behavior sequences.

Ax Haitong Ma, Ofir Nabati, Aviv Rosenberg, Bo Dai, Oran Lang, Craig Boutilier, Na Li, Shie Mannor, Lior Shani, Guy Tenneholtz 5/21/2026

Reinforcement Learning with Discrete Diffusion Policies for Combinatorial Action Spaces

Presents discrete diffusion models as policies for reinforcement learning with combinatorial action spaces using policy mirror descent for stable training.

Ax Yixuan Wang, Dan P. Guralnik, Saiedeh Akbari, Warren E. Dixon 5/21/2026

Effective Model Pruning: Measure The Redundancy of Model Components

Introduces Effective Model Pruning (EMP) framework using importance scores and effective sample size concept to determine optimal sparsity levels for neural networks.

Ax Linus Kreitner, Paul Hager, Jonathan Mengedoht, Georgios Kaissis, Daniel Rueckert, Martin J. Menten 5/21/2026

Efficient numeracy in language models through single-token number embeddings

Proposes single-token number embeddings to improve LLM numeracy and numerical reasoning without requiring external tools or excessive reasoning chains.

Ax Moritz Wagner, Christophe Roux, Max Zimmer, Sebastian Pokutta 5/21/2026

A Free Lunch in LLM Compression: Revisiting Retraining after Pruning

Research on post-training pruning for LLMs using local reconstruction to adapt weights after sparsity patterns are selected, reducing inference costs without full retraining.

Ax Shengfan Cao, Francesco Borrelli, Eunhyek Joa 5/21/2026

Constrained Policy Optimization via Sampling-Based Weight-Space Projection

Proposes SCPO, a sampling-based method for constrained policy optimization in safety-critical RL without gradient access to constraints.

Ax Abhisek Ganguly, Santosh Ansumali, Sauro Succi 5/21/2026

Deep Neural Networks as Discrete Dynamical Systems: Implications for Physics-Informed Learning

Analyzes relationship between deep neural networks and discrete dynamical systems, comparing PINN solutions to PDEs.

Ax Li Ju, Mayank Nautiyal, Andreas Hellander, Ekta Vats, Prashant Singh 5/21/2026

Epistemic Uncertainty Quantification for Pre-trained VLMs via Riemannian Flow Matching

REPVLM: quantifying epistemic uncertainty in vision-language models via Riemannian flow matching on embedding space hyperspheres.

Ax Yannis Montreuil, Letian Yu, Axel Carlier, Lai Xing Ng, Wei Tsang Ooi 5/21/2026

Learning to Defer in Non-Stationary Time Series via Switching State-Space Models

L2D-SLDS: online learning-to-defer framework for non-stationary time series using switching linear-Gaussian state-space models.

Ax Yuntong Hu, Yuting Zhang, Matthew Trager, Yi Zhang, Shuo Yang, Wei Xia, Stefano Soatto 5/21/2026

Evolutionary Generation of Multi-Agent Systems

Evolutionary generation of multi-agent system architectures using LLMs without code generation, enabling flexible and generalizable MAS design.

Ax Xian Wu, Kaijie Zhu, Ying Zhang, Lun Wang, Wenbo Guo 5/21/2026

rePIRL: Learn PRM with Inverse RL for LLM Reasoning

rePIRL: learning process reward models for LLM reasoning using inverse RL without requiring expert reward functions.

Ax Ning Yang, Chengzhi Wang, Yibo Liu, Baoliang Tian, Haijun Zhang 5/21/2026

CompilerKV: Risk-Adaptive KV Compression via Offline Experience Compilation

CompilerKV: risk-adaptive KV cache compression using offline experience compilation to make per-head reliability decisions across prompts.

Ax Jiwoo Hong, Shao Tang, Zhipeng Wang 5/21/2026

Bayesian Preference Learning for Test-Time Steerable Reward Models

ICRM: Bayesian method for learning in-context steerable reward models that adapt at test-time for complex preference distributions in LLM alignment.

Ax Andrei Kanavalau, Carmen Amo Alonso, Sanjay Lall 5/21/2026

Gated Normalization Removal and Scale Anchoring in Pre-Norm Transformers

TaperNorm gradually removes normalization layers in pre-norm transformers by transitioning to learned sample-independent maps.

Ax Ayse Tursucular, Ayush Mohanty, Nazal Mohamed, Nagi Gebraeel 5/21/2026

Federated Learning of Nonlinear Temporal Dynamics with Graph Attention-based Cross-Client Interpretability

Federated learning of nonlinear temporal dynamics with graph attention for understanding interdependent subsystems without sharing raw data.

Ax Xiuying Wei, Caglar Gulcehre 5/21/2026

RAT+: Train Dense, Infer Sparse -- Recurrence Augmented Attention for Dilated Inference

RAT+: training dense attention then inferring with structured dilate patterns reduces FLOPs and KV cache while preserving long-range connectivity.

Ax Jiayang Gao, Tianyi Zheng, Jiayang Zou, Fengxiang Yang, Shice Liu, Luyao Fan, Zheyu Zhang, Hao Zhang, Jinwei Chen, Peng-Tao Jiang, Bo Li, Jia Wang 5/21/2026

C$^2$FG: Control Classifier-Free Guidance via Score Discrepancy Analysis

Theoretical analysis of Classifier-Free Guidance with dynamic weight adjustment based on score discrepancy analysis for diffusion models.

Ax Ren Fujiwara, Yasuko Matsubara, Yasushi Sakurai 5/21/2026

When to Retrain after Drift: A Data-Only Test of Post-Drift Data Size Sufficiency

CALIPER: data-only test for detecting concept drift and estimating post-drift data size needed for stable model retraining.

Ax Tim Kaiser, Markus Kollmann 5/21/2026

Diffusion Models Memorize in Training -- and Generalize in Inference

Study of generalization vs memorization in diffusion models, showing training overfits denoising objective while maintaining sample-level generalization.

Ax Jeremy J Samuelson 5/21/2026

Informationally Compressive Anonymization: Non-Degrading Sensitive Input Protection for Privacy-Preserving Supervised Machine Learning

Privacy-preserving ML technique using informational compression anonymization to protect sensitive data without performance degradation.

Ax Zhenghang Song, Tang Qian, Lu Chen, Yushuai Li, Zhengke Hu, Bingbing Fang, Yumeng Song, Junbo Zhao, Sheng Zhang, Tianyi Li 5/21/2026

FEAT: A Linear-Complexity Foundation Model for Extremely Large Structured Data

Linear-complexity foundation model for structured data handling scalability challenges in healthcare, finance, and enterprise databases.

Ax Rohan Deb, Stephen J. Wright, Arindam Banerjee 5/21/2026

Inference Time Policy Optimization for Offline RL with Differentiable World Models

Offline reinforcement learning framework combining pretrained policies with learned world models for inference-time optimization.

Ax Victor Kawasaki-Borruat, Clara Grotehans, Pierre Vandergheynst, Adam Gosztolai 5/21/2026

Diffusion Processes on Implicit Manifolds

Mathematical formalism for constructing diffusion processes on lower-dimensional data manifolds using only point cloud samples.

Ax Yunhui Jang, Lu Zhu, Jake Fawkes, Alisandra Kaye Denton, Dominique Beaini, Emmanuel Noutahi 5/21/2026

Towards Autonomous Mechanistic Reasoning in Virtual Cells

Framework using LLMs for mechanistic reasoning in biology via structured explanations as action graphs for scientific discovery.

Ax Yuanda Xu, Hejian Sang, Zhengze Zhou, Ran He, Zhipeng Wang, Alborz Geramifard 5/21/2026

TIP: Token Importance in On-Policy Distillation

On-policy knowledge distillation method identifying which token positions provide most useful learning signals from teacher to student models.

Ax Qichuan Yin, Manzil Zaheer, Tian Li 5/21/2026

Differentially Private Model Merging

Post-processing techniques for model merging to satisfy differential privacy requirements without retraining across models with different privacy-utility tradeoffs.

Ax Adam Ousherovitch, Ambuj Tewari 5/21/2026

Compute Aligned Training: Optimizing for Test Time Inference

Compute Aligned Training aligns LLM post-training with test-time inference procedures that aggregate or filter outputs from scaled compute.

Ax Jerry Y. Huang, Justin Lin, Sheel Shah, Kartik Nair, Nicholas M. Boffi 5/21/2026

How to Guide Your Flow: Few-Step Alignment via Flow Map Reward Guidance

Flow-based guidance method reformulating reward maximization in generative models as deterministic optimal control for efficient few-step alignment.

Ax Yuanfan Li, Xiyuan Wei, Tianbao Yang, Yiming Ying 5/21/2026

Statistical Consistency and Generalization of Contrastive Representation Learning

Statistical consistency and generalization analysis of contrastive representation learning underlying modern foundation models.

Ax Jorge L. Ruiz Williams 5/21/2026

HeadQ: Model-Visible Distortion and Score-Space Correction for KV-Cache Quantization

HeadQ: KV-cache quantization method measuring error in model-visible attention coordinates rather than reconstruction error for improved inference.

Ax Moritz Grillo, Guido Mont\'ufar 5/21/2026

Most ReLU Networks Admit Identifiable Parameters

Theoretical analysis of when ReLU network parameters are identifiable beyond standard symmetries using weighted polyhedral complexes.

Ax Carlo Romeo, Girolamo Macaluso, Alessandro Sestini, Andrew D. Bagdanov 5/21/2026

SOPE: Stabilizing Off-Policy Evaluation for Online RL with Prior Data

Method for stabilizing off-policy evaluation in online reinforcement learning when incorporating prior data, eliminating manual tuning requirements.

Ax Yun Qu, Qi Wang, Yixiu Mao, Heming Zou, Yuhang Jiang, Yingyue Li, Wutong Xu, Lizhou Cai, Weijie Liu, Clive Bai, Kai Yang, Yangkun Chen, Saiyong Yang, Xiangyang Ji 5/21/2026

Listwise Policy Optimization: Group-based RLVR as Target-Projection on the LLM Response Simplex

Listwise policy optimization reveals geometric structure in group-based reinforcement learning for LLM post-training with verifiable rewards.

Ax Kartik Tandon, Julian Gould, Tanishq Bhatia, Francesca Dominici, Alejandro Ribeiro, Claudio Battiloro 5/21/2026

Consistent Geometric Deep Learning via Hilbert Bundles and Cellular Sheaves

Unified geometric deep learning framework for infinite-dimensional signals on irregular domains using Hilbert bundles and cellular sheaves.

Ax Chenyang Song, Weilin Zhao, Xu Han, Chaojun Xiao, Yingfa Chen, Zhiyuan Liu 5/21/2026

DECO: Sparse Mixture-of-Experts with Dense-Comparable Performance on End-Side Devices

DECO: sparse mixture-of-experts architecture designed for efficient edge device deployment with dense model performance and reduced memory footprint.

Ax Atul Ganju, Travis McVoy, Shaddin Dughmi, Shang-Hua Teng 5/21/2026

A Theory of Time-Sensitive Language Generation: Sparse Hallucination Beats Mode Collapse

Theoretical analysis of language generation under time-sensitive constraints, studying trade-offs between breadth and timeliness in string generation.

Ax Yuanda Xu, Hejian Sang, Zhengze Zhou, Ran He, Zhipeng Wang, Alborz Geramifard 5/21/2026

Beyond GRPO and On-Policy Distillation: An Empirical Sparse-to-Dense Reward Principle for Language-Model Post-Training

Research on reward allocation principles for LLM post-training, comparing sparse sequence-level vs dense token-level rewards for different model stages.

Ax Jack Young 5/21/2026

WriteSAE: Sparse Autoencoders for Recurrent State

WriteSAE introduces sparse autoencoders for matrix updates in recurrent language models like Mamba-2 and RWKV-7, learning rank-1 matrix atoms to directly replace model writes.

Ax Jiaqi Sun, Boyang Sun, Rasmy M. H., Xiangchen Song, Kun Zhang 5/21/2026

MoRe: Modular Representations for Principled Continual Representation Learning on Sequential Data

MoRe: modular representation learning approach for continual learning on sequential data with minimal catastrophic forgetting.

Ax Shukai Du, Junzhe Zhang, Yiming Li 5/21/2026

Lagrangian Flow Matching: A Least-Action Framework for Principled Path Design

Lagrangian flow matching framework for designing probability paths in generative models beyond standard straight-line transport.

Ax Yuan Zhang, Lifeng Guo, Junwen Pan, Wenzhao Zheng, Wen Zhou, Kuan Cheng, Kurt Keutzer, Shanghang Zhang 5/21/2026

SEED: Targeted Data Selection by Weighted Independent Set

SEED: data selection method formulated as weighted independent set problem to balance quality and diversity in training datasets.

Ax Yangyou Liu, Zezhi Shao, Xinyu Chen, Hu Chen, Fei Wang, Yuankai Wu 5/21/2026

PULSE: Generative Phase Evolution for Non-Stationary Time Series Forecasting

PULSE: generative model for non-stationary time series forecasting using phase evolution to handle distribution shifts.

Ax Yuhan Wang, Haopeng Zhang, Yibo Ding, Jiaqi Yu, Xinyu Zhao, Yuhang Liu, Ziwei Zhang, Xiao Wang, Ruijie Wang 5/21/2026

S2Aligner: Pair-Efficient and Transferable Pre-Training for Sparse Text-Attributed Graphs

Pre-training method for sparse text-attributed graphs using LLM alignment to improve transferability in sparse node text scenarios.

Ax Muhammad Umer, Muhammad Ahmed Mohsin, Ahsan Bilal, Arslan Chaudhry, Andreas Haupt, Sanmi Koyejo, Emily Fox, John M. Cioffi 5/21/2026

General Preference Reinforcement Learning

General Preference RL framework unifying online RL and preference optimization for LLM alignment across both structured and open-ended tasks.

Ax Weinuo Ou 5/21/2026

Exact Linear Attention

Exact Linear Attention mechanism achieving linear computational complexity for Transformers through kernel decomposition without approximation error.

Ax Yujie Lin, Chengyi Yang, Zhishang Xiang, Yiping Song, Jinsong Su 5/21/2026

ZeroUnlearn: Few-Shot Knowledge Unlearning in Large Language Models

Few-shot machine unlearning method for removing sensitive information from LLMs without retraining or degrading model utility.

Ax Han Guo, Jack Zhang, Arjun Menon, Driss Guessous, Vijay Thakkar, Yoon Kim, Tri Dao 5/21/2026

CODA: Rewriting Transformer Blocks as GEMM-Epilogue Programs

CODA optimization technique rewriting Transformer blocks as GEMM operations to reduce memory-bound operator bottlenecks in training.

Ax Lihong Lin, Haidong Kang 5/21/2026

Unlocking the Potential of Continual Model Merging: An ODE Perspective

Continual Model Merging framework using ODE perspective to control capacity allocation across sequential tasks, reducing catastrophic forgetting.

Ax David Pape, Jonathan Evertz, Lea Sch\"onherr 5/21/2026

The Silent Hyperparameter: Quantifying the Impact of Inference Backends on LLM Reproducibility

Study of how inference backend software systems impact LLM reproducibility and benchmark results at fractions of percentage point differences.