Ax Zhiyuan Zhai, Ming Li, Xin Wang 4/28/2026

Revisable by Design: A Theory of Streaming LLM Agent Execution

Proposes stream paradigm for LLM agents enabling concurrent user intervention during execution instead of transactional completion model.

Ax Varun Totakura, Ankita Singh, Yushun Dong, Shayok Chakraborty 4/28/2026

An Analysis of Active Learning Algorithms using Real-World Crowd-sourced Text Annotations

Analysis of active learning algorithms using crowdsourced annotations, examining performance when labeling oracles are imperfect in real-world settings.

Ax Marcel Hedman, Kale-ab Abebe Tessera, Juan Claude Formanek, Anya Sims, Riccardo Zamboni, Trevor McInroe, John Torr, Elliot Fosong 4/28/2026

CODA: Coordination via On-Policy Diffusion for Multi-Agent Offline Reinforcement Learning

CODA addresses coordination failure in multi-agent offline RL using diffusion-based trajectory generation for data augmentation to improve joint policy learning.

Ax Rory Young, Nicolas Pugeault 4/28/2026

GIFT: Global stabilisation via Intrinsic Fine Tuning

Deep RL policies for continuous control suffer from sensitivity to initial conditions. GIFT proposes intrinsic fine-tuning to stabilize state dynamics in nonlinear contact environments.

Ax Taihua Xu, Genhao Tian, Jicong Fan, Xibei Yang, Qinghua Zhang, Yun Cui 4/28/2026

Layer Embedding Deep Fusion Graph Neural Network

Layer Embedding Deep Fusion GNN addresses low-homophily graphs and long-range dependencies in graph neural networks.

Ax Liaoyaqi Wang, Chunsheng Zuo, William Jurayj, Benjamin Van Durme, Anqi Liu 4/28/2026

Process Supervision of Confidence Margin for Calibrated LLM Reasoning

RLCM framework uses calibration-aware reinforcement learning to prevent overconfidence in LLM reasoning with confidence margins.

Ax Hongtao Hao, Joseph L. Austerweil 4/28/2026

TEMPO: Transformers for Temporal Disease Progression from Cross-Sectional Data

TEMPO Transformer learns ordinal and continuous biomarker event sequences from cross-sectional medical data.

Ax Hanna R{\o}d, Dagny Streit, Nils Valseth Selte, Justin Li 4/28/2026

When Context Sticks: Studying Interference in In-Context Learning

Studies context stickiness interference in in-context learning where earlier prompt examples hinder adaptation to later tasks.

Ax Bingda Tang, Yuhui Zhang, Xiaohan Wang, Jiayuan Mao, Ludwig Schmidt, Serena Yeung-Levy 4/28/2026

V-GRPO: Online Reinforcement Learning for Denoising Generative Models Is Easier than You Think

V-GRPO framework uses online reinforcement learning to align denoising generative models with human preferences.

Ax Duc N. Do, Minh N. Do, Dang Nguyen, Khanh T. Q. Le, Khoa D. Pham, Hung N. Huynh, Phi Pham-Van-Hoang, Quan K. Huynh, Ramez M. Odat, Perisa Ashar, Ethan Philip Lowder, Minh H. N. Le, Hoang Le, Phat V. H. Nguyen, Quan Le, Jacques Kpodonu, Phat K. Huynh 4/28/2026

Domain-Adapted Fine-Tuning of ECG Foundation Models for Multi-Label Structural Heart Disease Screening

Fine-tuned ECG foundation models detect structural heart disease using multi-label learning on benchmark dataset.

Ax Tomer Zilca, Gal Mendelson 4/28/2026

Approximating Uniform Random Rotations by Two-Block Structured Hadamard Rotations in High Dimensions

Structured Hadamard rotations approximate uniform random rotations efficiently for embeddings, compression, and ML applications.

Ax Dikran Hovagimian 4/28/2026

Evolve: A Persistent Knowledge Lifecycle for Small Language Models

Evolve pairs small language models with persistent teacher-compiled knowledge stores refined via sleep consolidation.

Ax Lucky Verma 4/28/2026

When Does Removing LayerNorm Help? Activation Bounding as a Regime-Dependent Implicit Regularizer

Dynamic Tanh bounds activations as regime-dependent regularizer, replacing LayerNorm with mixed effects across model scales.

Ax Dan Liu, Fida K. Dankar, Jennifer C. deBruyn, Amanda Ricciuto, Anne M. Griffiths, Thomas D. Walters, Khaled EI Emam 4/28/2026

Machine learning models for estimating counterfactuals in a single-arm inflammatory bowel disease study

Machine learning models estimate counterfactual outcomes for single-arm clinical trials using external control data.

Ax Divakar Kumar Yadav, Tian Zhao, Deepak Kumar 4/28/2026

Evaluating CUDA Tile for AI Workloads on Hopper and Blackwell GPUs

Cross-architecture evaluation of NVIDIA's CUDA Tile for GPU kernel development on Hopper and Blackwell GPUs.

Ax Divakar Kumar Yadav, Tian Zhao 4/28/2026

Hybrid JIT-CUDA Graph Optimization for Low-Latency Large Language Model Inference

Hybrid JIT-CUDA Graph framework reducing LLM inference latency and kernel launch overhead for interactive applications.

Ax Audrey Cherilyn, Houman Safaai 4/28/2026

Supernodes and Halos: Loss-Critical Hubs in LLM Feed-Forward Layers

Analysis of loss-critical channels (supernodes) in transformer feed-forward layers showing concentrated importance in LLMs.

Ax Lichen Li, Hengguang Zhou, Yijun Liang, Tianyi Zhou, Cho-Jui Hsieh 4/28/2026

Do Synthetic Trajectories Reflect Real Reward Hacking? A Systematic Study on Monitoring In-the-Wild Hacking in Code Generation

Systematic study comparing synthetic reward hacking trajectories with naturally emerging hacking in code generation models using RL.

Ax Chen Zeng, Jiahui Wang, Qiao Wang 4/28/2026

Autocorrelation Reintroduces Spectral Bias in KANs for Time Series Forecasting

Theoretical study showing temporal autocorrelation reintroduces spectral bias in Kolmogorov-Arnold Networks for time series forecasting.

Ax Sifan Wang, Shawn Koohy, Yiping Lu, Paris Perdikaris 4/28/2026

When PINNs Go Wrong: Pseudo-Time Stepping Against Spurious Solutions

Analysis of physics-informed neural networks convergence failures and spurious solutions using pseudo-time stepping.

Ax Bingqing Jiang, Difan Zou 4/28/2026

On the Memorization of Consistency Distillation for Diffusion Models

Study of memorization vs generalization in consistency distillation for diffusion models during training.

Ax Rahul Narava, Siddharth Verma, Ojas Jain, Shashi Shekhar Jha, Mayank Shekhar Jha 4/28/2026

CAPSULE: Control-Theoretic Action Perturbations for Safe Uncertainty-Aware Reinforcement Learning

Safe reinforcement learning method using control theory to provide hard constraint-based safety guarantees in high-dimensional systems.

Ax Giansalvo Cirrincione 4/28/2026

Rank, Head-Channel Non-Identifiability, and Symmetry Breaking: A Precise Analysis of Representational Collapse in Transformers

Transformer representational collapse analysis: token rank convergence prevents pure self-attention; MLPs enable architectural diversity beyond prior conclusions.

Ax Antonij Mijoski, Marko Karbevski 4/28/2026

Can an MLP Absorb Its Own Skip Connection?

Analysis of when MLP skip connections absorb into residual-free architecture; homogeneous activations prevent absorption based on degree properties.

Ax Chenyi Li, Yanchen Nie, Zhengyu Ming, Gong Zhang, Kun Yuan, Zaiwen Wen 4/28/2026

OptProver: Bridging Olympiad and Optimization through Continual Training in Formal Theorem Proving

OptProver: formal theorem prover for optimization domain using continual training; addresses distribution shift between Olympiad and optimization problems.

Ax Viet-Hoang Tran, An Nguyen, Beno\^it Gu\'erand, Thieu N. Vo, Tan M. Nguyen 4/28/2026

Quasi-Equivariant Metanetworks

Quasi-Equivariant Metanetworks: architecture operating on pretrained weights accounting for parameter-function non-injectivity and inherent symmetries.

Ax Huadong Liao 4/28/2026

Transformer as an Euler Discretization of Score-based Variational Flow

Theoretical framework: Transformers as Euler discretization of Score-based Variational Flow for unified mathematical foundation of attention architecture.

Ax Alexis Limozin, Eduard Durech, Torsten Hoefler, Imanol Schlag, Valentina Pyatkin 4/28/2026

SFT-then-RL Outperforms Mixed-Policy Methods for LLM Reasoning

SFT-then-RL pipeline outperforms recent mixed-policy LLM reasoning methods; reveals bugs in DeepSpeed optimizer and baseline implementations.

Ax Shuaizhi Cheng, Xiang Shi, Mingwei Li 4/28/2026

The Override Gap: A Magnitude Account of Knowledge Conflict Failure in Hypernetwork-Based Instant LLM Adaptation

Hypernetwork-based LLM adaptation fails on knowledge conflicts due to magnitude scaling; adapter margin insufficient for document-pretraining contradictions.

Ax Mingze Li, Yu Rong, Songyou Li, Lihong Wang, Jiacheng Cen, Liming Wu, Anyi Li, Zongzhao Li, Qiuliang Liu, Rui Jiao, Tian Bian, Pengju Wang, Hao Sun, Jianfeng Zhang, Ji-Rong Wen, Deli Zhao, Shifeng Jin, Tingyang Xu, Wenbing Huang 4/28/2026

Agentic Fusion of Large Atomic and Language Models to Accelerate Materials Discovery

ElementsClaw: agentic framework combining Large Atomic Models with LLMs for autonomous materials discovery orchestration.

Ax Chih-Chung Hsu, Xin-Di Ma, Wo-Ting Liao, Chia-Ming Lee 4/28/2026

ELSA: Exact Linear-Scan Attention for Fast and Memory-Light Vision Transformers

ELSA algorithm for efficient attention in vision transformers with exact softmax semantics and O(u log n) error bound.

Ax Aditi De 4/28/2026

Symmetric Equilibrium Propagation for Thermodynamic Diffusion Training

Analog substrate implementation of equilibrium propagation for thermodynamic diffusion model training with energy efficiency gains.

Ax Zhengding Hu, Hehua Ouyang, Chang Chen, Zaifeng Pan, Yue Guan, Zhongkai Yu, Zhen Wang, Steven Swanson, Yufei Ding 4/28/2026

JigsawRL: Assembling RL Pipelines for Efficient LLM Post-Training

Framework for cost-efficient LLM post-training via RL with pipeline multiplexing and dynamic resource allocation across workers.

Ax Jonathan Hoss, Moritz Link, Noah Klarmann 4/28/2026

Scalable Production Scheduling: Linear Complexity via Unified Homogeneous Graphs

RL-based policy for job shop scheduling using unified homogeneous graphs to achieve linear complexity and topological robustness.

Ax Nicola Zanarini, Niccol\`o Ferrari 4/28/2026

Graph Memory Transformer (GMT)

Transformer architecture replacing feed-forward layers with learned memory graphs for efficient routing while maintaining causal self-attention.

Ax Niels Bracher, Xavier Intes, Stefan T. Radev 4/28/2026

Inverting Foundation Models of Brain Function with Simulation-Based Inference

Method pairing brain activity foundation models with LLMs to invert neural responses and recover stimuli from synthetic brain activity.

Ax Valerie Tsao, Nathaniel Chaney, Manolis Veveakis 4/28/2026

Learning Interpretable PDE Representations for Generative Reconstructions with Structured Sparsity

Latent diffusion framework combining structured sparsity and PDE representations for sparse field reconstruction and super-resolution.

Ax Sinjini Mitra, Constantine Kyriakakis, Shenyuan Liang, Anuj Srivastava, Pavan Turaga 4/28/2026

Geometry Preserving Loss Functions Promote Improved Adaptation of Blackbox Generative Model

Loss functions preserving geometric properties for adapting blackbox generative models without access to weights or architecture.

Ax Ibne Farabi Shihab, Sanjeda Akter, Anuj Sharma 4/28/2026

Continual Calibration: Coverage Can Collapse Before Accuracy in Lifelong LLM Fine-Tuning

Analysis of uncertainty calibration degradation in continual learning for LLMs, showing coverage collapse occurs before accuracy loss in sequential fine-tuning.

Ax Yanwei Cui, Guanghui Wang, Xing Zhang, Peiyang He, Ziyuan Li, Bing Zhu, Wei Qiu, Xusheng Wang, Zheng Yu, Anqi Xin 4/28/2026

Hindsight Preference Optimization for Financial Time Series Advisory

Method applying reinforcement learning preference alignment to train LLMs for financial time series advisory with actionable signals and risk management.

Ax Yuto Tanaka, Issei Sato 4/28/2026

Fix Initial Codes and Iteratively Refine Textual Directions Toward Safe Multi-Turn Code Correction

Framework for multi-turn code correction using LLMs with iterative refinement of textual directions, building on Monte Carlo Tree Search approaches.

Ax Danny Wang, Ruihong Qiu, Zi Huang 4/28/2026

When to Commit? Towards Variable-Size Self-Contained Blocks for Discrete Diffusion Language Models

Study of discrete diffusion language models addressing training-inference mismatch in blockwise semi-autoregressive decoding with variable-size token commitments.

Ax Jiaqi Wang, Wenhao Zhang, Weijie Shi, Yaliang Li, James Cheng 4/28/2026

TCOD: Exploring Temporal Curriculum in On-Policy Distillation for Multi-turn Autonomous Agents

Research on improving on-policy distillation for multi-turn AI agents by addressing trajectory-level KL instability when transferring reasoning from larger to smaller models.

Ax Ibne Farabi Shihab, Sanjeda Akter, Anuj Sharma 4/28/2026

Coverage-Based Calibration for Post-Training Quantization via Weighted Set Cover over Outlier Channels

Post-training quantization method for LLMs using weighted set cover to select calibration samples that activate outlier channels for better compression.

Ax Yutong He, Zhengyang Huang, Jiahe Geng 4/28/2026

FedSLoP: Memory-Efficient Federated Learning with Low-Rank Gradient Projection

FedSLoP algorithm reduces communication and memory costs in federated learning via low-rank gradient projections for resource-constrained environments.

Ax Rezaul Karim, Austin Wen, Wang Zongzuo, Weiwei Zhang, Yang Liu, Walid Ahmed 4/28/2026

FlashOverlap: Minimizing Tail Latency in Communication Overlap for Distributed LLM Training

FlashOverlap minimizes tail latency in communication-computation overlap for distributed LLM training.

Ax Zean Han, Ruihan Lin, Zezhen Ding, Jiheng Zhang 4/28/2026

Geometry-Aware Offline-to-Online Learning in Linear Contextual Bandits

Ellipsoidal confidence regions enable geometry-aware offline-to-online learning in contextual bandits.

Ax Jun Shu, Junxiong Jia, Deyu Meng, Zongben Xu 4/28/2026

A Limit Theory of Foundation Models: A Mathematical Approach to Understanding Emergent Intelligence and Scaling Laws

Limit theory formalizes emergent intelligence and scaling laws in foundation models mathematically.

Ax Hojoon Kim, Yuheng Wu, Thierry Tambe 4/28/2026

AgenticCache: Cache-Driven Asynchronous Planning for Embodied AI Agents

AgenticCache framework reduces LLM calls for embodied AI agents by caching and reusing plans.

Ax Wenjie Du, Yiyuan Yang, Tianxiang Zhan, Qingsong Wen 4/28/2026

End-to-End Learning for Partially-Observed Time Series with PyPOTS

PyPOTS open-source toolkit for end-to-end ML on partially-observed time series with missing values.