Isolater - Feed

Ax Haolong Chen, Liang Zhang, Zhengyuan Xin, Guangxu Zhu 5/25/2026

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

Multiscale Mamba-based mixture model for long-term spatio-temporal dependencies in time series prediction.

Ax Songyu Ke, Chenyu Wu, Yuxuan Liang, Huiling Qin, Junbo Zhang, Yu Zheng 5/25/2026

GeoMAE: Masking Representation Learning for Spatio-Temporal Graph Forecasting with Missing Values

Masked representation learning approach for spatio-temporal forecasting with missing data in urban systems.

Ax Quan Zhou, Jakub Marecek, Robert Shorten 5/25/2026

Online Learning with Multiple Fairness Regularizers via Graph-Structured Feedback

Bandit algorithm for learning multiple competing fairness objectives adaptively with graph-structured feedback.

Ax William Xu, Chenyu Zhang, Yihan Wang, Matthew Y. R. Yang, Zuoqiu Liu, Gautam Kamath, Yaoliang Yu, Yiwei Lu 5/25/2026

Are Targeted Data Poisoning Attacks as Effective as We Think?

Analysis of data poisoning attack effectiveness on hardest-to-poison samples versus average attack success rates.

Ax Liran Nochumsohn, Raz Marshanski, Hedi Zisling, Omri Azencot 5/25/2026

Super-Linear: A Lightweight Pretrained Mixture of Linear Experts for Time Series Forecasting

Lightweight mixture-of-experts model for time series forecasting with improved computational efficiency over large pretrained models.

Ax Xin-Qiang Cai, Wei Wang, Feng Liu, Tongliang Liu, Gang Niu, Masashi Sugiyama 5/25/2026

Reinforcement Learning with Verifiable yet Noisy Rewards under Imperfect Verifiers

RL framework addressing imperfect verifier noise through stochastic reward channels with asymmetric false positive/negative rates.

Ax Lunjun Zhang, Shuo Han, Hanrui Lyu, Bradly C Stadie 5/25/2026

D2 Actor Critic: Diffusion Actor Meets Distributional Critic

Diffusion actor-critic RL algorithm with distributional critic for training expressive policies without policy gradient variance.

Ax Weishuo Ma, Yanbo Wang, Xiyuan Wang, Lei Zou, Muhan Zhang 5/25/2026

GILT: An LLM-Free, Tuning-Free Graph Foundational Model for In-Context Learning

Graph foundational model enabling in-context learning on heterogeneous graphs without LLMs or tuning.

Ax Aanchal Rajesh Chugh, Marion Neumeier, Sebastian Dorn 5/25/2026

X-TRACK: Physics-Aware xLSTM for Realistic Vehicle Trajectory Prediction

Physics-aware xLSTM architecture for vehicle trajectory prediction in autonomous driving, addressing LSTM memory limitations.

Ax Jayden Teoh, Manan Tomar, Kwangjun Ahn, Edward S. Hu, Tim Pearce, Pratyusha Sharma, Akshay Krishnamurthy, Riashat Islam, Alex Lamb, John Langford 5/25/2026

Next-Latent Prediction Transformers Learn Compact World Models

NextLat: Transformer training with self-supervised next-latent prediction to learn compact world models with better generalization.

Ax Mahdi Mohammadigohari, Giuseppe Di Fatta, Giuseppe Nicosia, Panos M. Pardalos 5/25/2026

Operator-Based Generalization Bound for Deep Learning: Insights on Multi-Task Learning

Operator-theoretic framework for generalization bounds in multi-task learning with improved tightness via Koopman-based approaches.

Ax Mahdi Mohammadigohari, Giuseppe Di Fatta, Giuseppe Nicosia, Panos M. Pardalos 5/25/2026

On the Koopman-Based Generalization Bounds for Multi-Task Deep Learning

Generalization bounds for multitask deep learning using Koopman operator theory with tighter guarantees than norm-based methods.

Ax Guikun Xu, Xiaohan Yi, Ziqiao Meng, Peilin Zhao, Yatao Bian 5/25/2026

Energy-Guided Generative Modeling for Low-Energy Molecular Structure Discovery

Generative model for discovering low-energy molecular conformations by combining energy guidance with learned representations.

Ax Weijie Shi, Yanxi Chen, Zexi Li, Xuchen Pan, Yuchang Sun, Jiajie Xu, Xiaofang Zhou, Yaliang Li 5/25/2026

R$^3$L: Reflect-then-Retry Reinforcement Learning with Language-Guided Exploration, Pivotal Credit, and Positive Amplification

R3L: RL framework for LLM reasoning and agents using language-guided exploration, credit assignment, and positive amplification for improved performance.

Ax Jun Liu, Leo Yu Zhang, Fengpeng Li, Isao Echizen, Jiantao Zhou 5/25/2026

Low-Cost Hard-Label Adversarial Attack with Theoretical Foundations

Hard-label adversarial attack method with theoretical foundations addressing initialization and optimization for black-box threat models.

Ax Wei Lin, Yining Jiang, Qingyu Song, Qiao Xiang, Hong Xu 5/25/2026

AGZO: Activation-Guided Zeroth-Order Optimization for LLM Fine-Tuning

AGZO: Zeroth-order optimization for LLM fine-tuning using activation structure guidance to reduce memory constraints without backpropagation.

Ax Wei-Di Chang, Mikael Henaff, Brandon Amos, Gregory Dudek, Scott Fujimoto 5/25/2026

The Surprising Difficulty of Search in Model-Based Reinforcement Learning

Analysis showing search can harm model-based RL performance even with accurate models; overestimation bias mitigation is more critical than prediction accuracy.

Ax Tim Tomov, Dominik Fuchsgruber, Stephan G\"unnemann 5/25/2026

Task-Awareness Improves LLM Generations and Uncertainty

Task-aware LLM decoding that models outputs in task-dependent latent structures for better generation quality and uncertainty estimation.

Ax Eloi Campagne (CB), Yvenn Amara-Ouali (LMO), Yannig Goude (LMO), Mathilde Mougeot (CB, ENSIIE, ENS Paris Saclay), Argyris Kalogeratos (CB, ENS Paris Saclay) 5/25/2026

Cascaded Transfer: Learning Many Tasks under Budget Constraints

Cascaded Transfer Learning paradigm for learning many related tasks hierarchically under distributed training budget constraints.

Ax Silas Ruhrberg Est\'evez, Christopher Chiu, Mihaela van der Schaar 5/25/2026

Automatic Construction of Clinical Scoring Systems with LLM Agents

LLM agents automatically construct interpretable clinical scoring systems that align with workflow constraints like memorability and auditability.

Ax Tian Zhang, Yujia Tong, Junhao Dong, Ke Xu, Yuze Wang, Jingling Yuan 5/25/2026

Forget by Uncertainty: Orthogonal Entropy Unlearning for Quantized Neural Networks

Orthogonal entropy unlearning method for removing memorized data from quantized neural networks while maintaining performance. GDPR-compliant.

Ax Jonathan N\"other, Adish Singla, Goran Radanovic 5/25/2026

MaMa: A Game-Theoretic Approach for Designing Safe Agentic Systems

MaMa: Game-theoretic framework for designing safe multi-agent systems that remain secure when individual LLM agents are compromised or adversarial.

Ax Atefeh Gilani, Sajani Vithana, Carol Xuan Long, Oliver Kosut, Lalitha Sankar, Flavio P. Calmon 5/25/2026

ArcMark: Distortion-Free Multi-Byte LLM Watermark via Optimal Transport

ArcMark: Multi-byte watermarking technique for LLM outputs using optimal transport to encode information without distorting text generation.

Ax Francesco Innocenti, El Mehdi Achour, Rafal Bogacz 5/25/2026

On the Infinite Width and Depth Limits of Predictive Coding Networks

Theoretical analysis of predictive coding networks at infinite width/depth limits. Biologically plausible alternative to backpropagation.

Ax Alina Shutova, Alexandra Olenina, Ivan Vinogradov, Anton Sinitsin 5/25/2026

Evaluating Memory Structure in LLM Agents

Benchmark for evaluating long-term memory architectures in LLM agents. Analyzes memory design capabilities for agent systems.

Ax Dongxiao He, Wenxuan Sun, Yongqi Huang, Jitao Zhao, Di Jin 5/25/2026

GP2F: Cross-Domain Graph Prompting with Adaptive Fusion of Pre-trained Graph Neural Networks

Cross-domain graph prompting framework adapting pre-trained GNNs. Graph neural network transfer learning for domain shift.

Ax Xin-Qiang Cai, Masashi Sugiyama 5/25/2026

VI-CuRL: Stabilizing Verifier-Independent RL Reasoning via Confidence-Guided Variance Reduction

VI-CuRL stabilizes verifier-free RL training for LLM reasoning via confidence-guided variance reduction. Enables scalable reinforcement learning without external verifiers.

Ax Hanna Benarroch (DI-ENS), Jamal Atif (CMAP), Olivier Capp\'e (DI-ENS) 5/25/2026

Certified Per-Instance Unlearning Using Individual Sensitivity Bounds

Per-instance noise calibration for certified machine unlearning with differential privacy. Improves unlearning efficiency via adaptive sensitivity bounds.

Ax Thanh Q. Tran, Arun Verma, Kiwan Wong, Bryan Kian Hsiang Low, Daniela Rus, Wei Xiao 5/25/2026

BarrierSteer: LLM Safety via Learning Barrier Steering

BarrierSteer inference-time safety framework for LLMs against adversarial attacks. Novel safety mechanism with theoretical grounding.

Ax L. Juli\'an Lechuga L\'opez, Farah E. Shamout, Tim G. J. Rudner 5/25/2026

An Empirical Analysis of Calibration and Selective Prediction in Multimodal Clinical Condition Classification

Calibration and selective prediction in multimodal clinical AI. Safety evaluation for medical AI systems.

Ax Lukas Thede, Stefan Winzeck, Zeynep Akata, Jonathan Richard Schwarz 5/25/2026

CapTrack: Multifaceted Evaluation of Forgetting in LLM Post-Training

CapTrack evaluates forgetting in LLM post-training beyond accuracy metrics. Comprehensive benchmark for catastrophic forgetting in LLMs.

Ax Woogyeol Jin, Taywon Min, Yongjin Yang, Swanand Ravindra Kadhe, Yi Zhou, Dennis Wei, Nathalie Baracaldo, Kimin Lee 5/25/2026

Entropy-Aware On-Policy Distillation of Language Models

Entropy-aware on-policy distillation for knowledge transfer between language models. Improves student LLM training stability and diversity.

Ax Zongyu Guo, Jiajun He, Zhaoyang Jia, Xiaoyi Zhang, Jiahao Li, Xiao Li, Bin Li, Jos\'e Miguel Hern\'andez-Lobato, Yan Lu 5/25/2026

Compression as Adaptation: Implicit Visual Representation with Diffusion Foundation Models

Diffusion foundation models as implicit visual representations with low-rank adaptations. Novel representation framework for vision models.

Ax Tianyu Pang, Yujie Fang, Zihang Liu, Shenyang Deng, Lei Hsiung, Shuhua Yu, Yaoqing Yang 5/25/2026

HTMuon: Improving Muon via Heavy-Tailed Spectral Correction

HTMuon optimizer improves Muon training for LLMs via heavy-tailed spectral correction. Novel optimization method for LLM training.

Ax Hoang Phan, Quang H. Nguyen, Hung T. Q. Le, Xiusi Chen, Heng Ji, Khoa D. Doan 5/25/2026

Decoding the Critique Mechanism in Large Reasoning Models

Analysis of critique mechanisms in Large Reasoning Models, investigating how models recover from errors through self-verification and backtracking.

Ax Tianyang Luo, Tao Feng, Zhigang Hua, Yan Xie, Shuang Yang, Ge Liu, Jiaxuan You 5/25/2026

MemReward: Graph-Based Experience Memory for LLM Reward Prediction with Limited Labels

MemReward framework using graph-based experience memory to enable LLM reward prediction with limited labels for reasoning task reinforcement learning.

Ax Ting Hu, Luanda Cai, Manolis Vlatakis 5/25/2026

Learning Safely Without Knowing the World:COMPASS-Hedge

COMPASS-Hedge algorithm balancing regret guarantees across adversarial and stochastic settings with baseline safety guarantees in online learning.

Ax Chenglin Li, Grant Ruan, Hua Geng 5/25/2026

Safe Reinforcement Learning with Preference-based Constraint Inference

Safe reinforcement learning framework using preference-based constraint inference for safety-critical decision making with minimal expert demonstrations.

Ax Jing Sun 5/25/2026

Representation over Routing: Overcoming Surrogate Hacking in Multi-Timescale PPO

Research on multi-timescale PPO with temporal credit assignment in reinforcement learning, addressing surrogate hacking through representation learning.

Ax Xiaoyang Liu, Zineng Dong, Yifan Bai, Yantao Li, Yuntian Liu, Tao Luo 5/25/2026

Decompose, Structure, and Repair: A Neuro-Symbolic Framework for Autoformalization via Operator Trees

Neuro-symbolic framework using operator trees for autoformalization, translating natural language math problems to formal language via hierarchical LLM reasoning.

Ax Grigorios Papanikolaou, Ioannis Kontopoulos, Konstantinos Tserpes 5/25/2026

A Comparative Analysis on the Performance of Upper Confidence Bound Algorithms in Adaptive Deep Neural Networks

Research on Upper Confidence Bound algorithms in Adaptive Deep Neural Networks for energy-efficient edge computing inference with dynamic accuracy-latency tradeoffs.

Ax Sajjad Khan 5/25/2026

S-Bus: Automatic Read-Set Reconstruction for Multi-Agent LLM State Coordination

S-Bus: HTTP middleware for concurrency control in multi-agent LLM systems with automatic read-set reconstruction.

Ax Pei Yang, Wanyi Chen, Tongyun Yang, Pengbin Feng, Jiarong Xing, Wentao Guo, Yuhang Yao, Yuhang Han, Hanchen Li, Xu Wang, Zeyu Wang, Jie Xiao, Anjie Yang, Liang Tian, Lynn Ai, Eric Yang, Tianyu Shi 5/25/2026

TwinRouterBench: Fast Static and Live Dynamic Evaluation for Realistic Agentic LLM Routing

TwinRouterBench: benchmark for evaluating LLM routing in agentic systems like coding agents and computer-use agents.

Ax Thomas Sommariva, Francesca Morandi, Simone Calderara, Angelo Porrello 5/25/2026

Distilling Linearized Behavior into Non-Linear Fine-Tuning for Effective Task Arithmetic

Method for distilling linearized fine-tuning behavior into non-linear models for effective task vector composition.

Ax Xiaocan Li, Shiliang Wu, Zheng Shen 5/25/2026

Decomposing MXFP4 quantization error for LLM reinforcement learning: reducible bias, recoverable deadzone, and an irreducible floor

Analysis of MXFP4 quantization error decomposition for LLM reinforcement learning post-training acceleration.

Ax Emma Leonhart 5/25/2026