Isolater - Feed

Ax Xiaofan Xia, Qin Li, Wenlong Mou 5/7/2026

Provable imitation learning for control of instability in partially-observed Vlasov--Poisson equations

Imitation learning approach for plasma stabilization control in Vlasov-Poisson systems with partial state observability constraints.

Ax Andrea Napoli, Paul White 5/7/2026

Order Matters: Improving Domain Adaptation by Reordering Data

ORDERED algorithm for unsupervised domain adaptation that reduces variance in discrepancy estimation through optimal data reordering.

Ax Andreas Pattichis, Constantine Dovrolis 5/7/2026

Continual Knowledge Updating in LLM Systems: Learning Through Multi-Timescale Memory Dynamics

Memini system for continual knowledge updating in LLMs using multi-timescale memory dynamics inspired by biological memory mechanisms.

Ax Harin Lee, Min-hwan Oh 5/7/2026

Unified Framework of Distributional Regret in Multi-Armed Bandits and Reinforcement Learning

Theoretical framework for analyzing regret distribution in multi-armed bandits and episodic RL with probabilistic guarantees across confidence levels.

Ax Tianshu Zhu, Wenyu Zhang, Xiaoying Zuo, Lun Tian, Haotian Zhao, Yucheng Zeng, Jingnan Gu, Daxiang Dong, Jianmin Wu, Dawei Yin, Dou Shen 5/7/2026

Rollout Pass-Rate Control: Steering Binary-Reward RL Toward Its Most Informative Regime

RL method for improving sample efficiency in agentic tasks by controlling pass-rate to optimal 50% for maximizing reward signal informativeness.

Ax Mariia Seleznova 5/7/2026

How Long Does Infinite Width Last? Signal Propagation in Long-Range Linear Recurrences

Theoretical study of signal propagation in finite-width linear recurrent models, analyzing approximation accuracy as sequence depth and width grow jointly.

Ax Marco Rando, Samuel Vaiter 5/7/2026

On the Hardness of Junking LLMs

Analysis of jailbreak attack difficulty on LLMs, showing random search effectiveness challenges assumptions about prompt structure necessity.

Ax Dan Wilson, Mohamed Akrout 5/7/2026

Low-Cost Black-Box Detection of LLM Hallucinations via Dynamical System Prediction

Low-cost black-box method to detect LLM hallucinations by treating model as dynamical system and analyzing embedding manifolds.

Ax Freyaa Chawla, Ahan Chawla, Rishi Singh, Joe Germino, Grigorii Khvatskii 5/7/2026

Human-AI Co-Mentorship in Project-Based Learning: A Case Study in Financial Forecasting

Case study of AI tools assisting high school students in financial forecasting research project with human-AI co-mentorship.

Ax Alper Y{\i}ld{\i}r{\i}m 5/7/2026

Superposition Is Not Necessary: A Mechanistic Interpretability Analysis of Transformer Representations for Time Series Forecasting

Mechanistic interpretability study of transformer time series forecasting using sparse autoencoders, questioning superposition hypothesis.

Ax Alexander Hsu, Zhaiming Shen, Wenjing Liao, Rongjie Lai 5/7/2026

Understanding In-Context Learning for Nonlinear Regression with Transformers: Attention as Featurizer

Theoretical analysis of transformer in-context learning for nonlinear regression, explaining how attention acts as a featurizer.

Ax Wilson Wu, Victor Lecomte, Michael Winer, George Robinson, Jacob Hilton, Paul Christiano 5/7/2026

Estimating the expected output of wide random MLPs more efficiently than sampling

Method to estimate expected output of wide random MLPs without sampling by computing activation distributions analytically.

Ax Kwan Soo Shin 5/7/2026

The Reasoning Trap: An Information-Theoretic Bound on Closed-System Multi-Step LLM Reasoning

Information-theoretic analysis showing multi-agent LLM debate preserves answer accuracy but degrades reasoning quality (Reasoning Trap).

Ax Yiming Huang, Zhenbo Shi, Xin-Cheng Wen, Jichuan Zeng, Cuiyun Gao, Peiyi Han, Chuanyi Liu 5/7/2026

Free Energy-Driven Reinforcement Learning with Adaptive Advantage Shaping for Unsupervised Reasoning in LLMs

FREIA: Free energy-driven reinforcement learning with adaptive advantage shaping for unsupervised LLM self-improvement and reasoning.

Ax Yiming Huang, Zhenbo Shi, Shuzheng Gao, Cuiyun Gao, Peiyi Han, Chuanyi Liu 5/7/2026

Adapt to Thrive! Adaptive Power-Mean Policy Optimization for Improved LLM Reasoning

Adaptive Power-Mean Policy Optimization method for improving LLM reasoning through adaptive reinforcement learning with verifiable rewards.

Ax Vageesh Kumar Saxena 5/7/2026

Connecting online criminal behavior with machine learning: Using authorship attribution to analyze and link potential online traffickers

arXiv work using authorship attribution machine learning to link anonymous online criminal profiles in trafficking networks.

Ax Beomchang Kim, Subhrajyoty Roy, Priyam Das 5/7/2026

BOOOM: Loss-Function-Agnostic Black-Box Optimization over Orthonormal Manifolds for Machine Learning and Statistical Inference

BOOOM: black-box optimization method over orthonormal manifolds for non-convex and derivative-free machine learning problems.

Ax Tejas Kulkarni, Antti Koskela, Laith Zumot 5/7/2026

Membership Inference Attacks for Retrieval Based In-Context Learning for Document Question Answering

arXiv security research on membership inference attacks against retrieval-augmented in-context learning in document QA systems.

Ax Emil Sharafutdinov, Ingemar Andr\'e 5/7/2026

Tree-Conditioned Edit Flows for Ancestral Sequence Reconstruction

arXiv paper on tree-conditioned edit flows for ancestral protein sequence reconstruction using deep learning.

Ax Lin Song, Wenbo Li, Guoqing Ma, Wei Tang, Bo Wang, Yuan Zhang, Yijun Yang, Yicheng Xiao, Jianhui Liu, Yanbing Zhang, Guohui Zhang, Wenhu Zhang, Hang Xu, Nan Jiang, Xin Han, Haoze Sun, Maoquan Zhang, Haoyang Huang, Nan Duan 5/7/2026

Awaking Spatial Intelligence in Unified Multimodal Understanding and Generation

JoyAI-Image: unified multimodal foundation model combining spatially enhanced MLLM with diffusion transformer for image understanding and generation.

Ax Hoffmann Muki, Olukunle Owolabi 5/7/2026

Are LLMs Ready for Conflict Monitoring? Empirical Evidence from West Africa

Evaluation of open-weight LLMs (Gemma, Llama, Mistral, OLMo) on conflict monitoring for West Africa against ACLED benchmark.

Ax Iman Sharifi, Peng Wei, Saber Fallah 5/7/2026

ANDRE: An Attention-based Neuro-symbolic Differentiable Rule Extractor

ANDRE: neuro-symbolic approach combining attention mechanisms with logic programming for interpretable rule extraction from data.

Ax Sarthak Choudhary, Atharv Singh Patlan, Nils Palumbo, Ashish Hooda, Kassem Fawaz, Somesh Jha 5/7/2026

Undetectable Backdoors in Model Parameters: Hiding Sparse Secrets in High Dimensions

arXiv paper on supply-chain backdoor attacks hiding sparse perturbations in image classifiers and Vision Transformers.

Ax Haruto Nakashima, Siddhartha Ganguly, Kenji Kashima 5/7/2026

Globally Solving Unbalanced Optimal Transport and Density Control for Gaussian Distributions

arXiv theoretical work on unbalanced optimal transport and density control for Gaussian distributions.

Ax Alessandro Micheli, Silvia Sapora, Anthea Monod, Samir Bhatt 5/7/2026

Entropic Riemannian Neural Optimal Transport

arXiv paper on optimal transport methods for curved spaces and Riemannian manifolds in machine learning.

Ax Jie Zhang, Pura Peetathawatchai, Florian Tram\`er, Avital Shafran 5/7/2026

Laundering AI Authority with Adversarial Examples

arXiv research on adversarial examples against vision-language models for fact-checking and content moderation.

Ax Anju Rani, Daniel Ortiz-Arroyo, Petar Durdevic 5/7/2026

Imagery Dataset for Remaining Useful Life Estimation of Synthetic Fibre Ropes

arXiv dataset paper on synthetic fiber rope degradation for remaining useful life estimation using imagery.

Ax St\'ephane d'Ascoli, J\'er\'emy Rapin, Yohann Benchetrit, Teon Brooks, Katelyn Begany, Jos\'ephine Raugel, Hubert Banville, Jean-R\'emi King 5/7/2026

A foundation model of vision, audition, and language for in-silico neuroscience

TRIBE v2: tri-modal foundation model predicting human brain activity from video, audio, language across 720 fMRI subjects.

Ax Zetai Cen, Jin Zhu, Xinwei Shen, Chengchun Shi 5/7/2026

Perturbation is All You Need for Extrapolating Language Models

Perturbation-based training framework for LLMs using semantic neighbor prefixes instead of exact tokens for improved extrapolation.

Ax Yixuan Mei, Zikun Li, Zixuan Chen, Shiqi Pan, Mengdi Wu, Xupeng Miao, Zhihao Jia, K. V. Rashmi 5/7/2026

Coral: Cost-Efficient Multi-LLM Serving over Heterogeneous Cloud GPUs

Coral: multi-LLM inference serving framework optimizing heterogeneous GPU allocation across multiple models with adaptive scheduling.

Ax Zhenhan Huang, Pin-Yu Chen, Tejaswini Pedapati, Jianxi Gao 5/7/2026

Intermediate Representations are Strong AI-Generated Image Detectors

AI-generated image detection using intermediate representations from vision models without retraining.

Ax Wenjun Yu, Shuguang Han, Amelie Chi Zhou 5/7/2026

One Pool, Two Caches: Adaptive HBM Partitioning for Accelerating Generative Recommender Serving

Coral: adaptive multi-LLM serving system for heterogeneous cloud GPUs with dynamic resource allocation across embedding and KV caches.

Ax Varad Vishwarupe, Nigel Shadbolt, Marina Jirotka, Ivan Flechais 5/7/2026

Deployment-Relevant Alignment Cannot Be Inferred from Model-Level Evaluation Alone

Analysis that model-level LLM alignment evaluation is insufficient; deployment-relevant alignment requires system-level evaluation.

Ax Jingtao Zhou, Xirui Kang, Feiyang Huang, Lai-Man Po 5/7/2026

SpecPL: Disentangling Spectral Granularity for Prompt Learning

SpecPL: prompt learning method for vision-language models using spectral granularity and counterfactual supervision.

Ax Ivan Bondarenko, Roman Derunets, Oleg Sedukhin, Mikhail Komarov, Ivan Chernov, Mikhail Kulakov 5/7/2026

RaguTeam at SemEval-2026 Task 8: Meno and Friends in a Judge-Orchestrated LLM Ensemble for Faithful Multi-Turn Response Generation

SemEval-2026 winning system: ensemble of 7 LLMs with GPT-4o judge for multi-turn faithful response generation with reference passages.

Ax Yepeng Weng, Qiao Hu, Takehisa Yairi 5/7/2026

UniVer: A Unified Perspective for Multi-step and Multi-draft Speculative Decoding

UniVer framework unifying multi-step and multi-draft speculative decoding for LLMs via optimal transport verification.

Ax Cyril Allauzen, Tom Bagby, Georg Heigold, Ehsan Variani, Ke Wu 5/7/2026

Benchmarking LLMs on the Massive Sound Embedding Benchmark (MSEB)

Evaluation of LLMs on audio embedding benchmark (MSEB), testing multimodal capabilities of models like Gemini on audio understanding tasks.

Ax Xiao Wang, Yifei Zhang, YongKang Liu, Xiaocui Yang, Zihan Wang, Shi Feng, Daling Wang 5/7/2026

From Parameter Dynamics to Risk Scoring : Quantifying Sample-Level Safety Degradation in LLM Fine-tuning

Research on safety degradation in LLM fine-tuning, analyzing parameter dynamics during training and quantifying sample-level safety risks.

Ax Zheng Fang, Xiaosen Wang, Shenyi Zhang, Shaokang Wang, Zhijin Ge 5/7/2026

Sparse Tokens Suffice: Jailbreaking Audio Language Models via Token-Aware Gradient Optimization

Token-aware gradient optimization attack on audio language models exploiting non-uniform gradient distribution for efficient jailbreaking.

Ax Kang Liu, Wei Peng, Jianchen Hu 5/7/2026

Budget-aware Auto Optimizer Configurator

Budget-Aware Optimizer Configurator (BAOC) that reduces GPU memory usage by assigning differentiated optimizer states across network blocks.

Ax Venkat Srinivasan, Vishaal Jatav, Anushka Chandrababu, Geetika Sharma 5/7/2026

Gyan: An Explainable Neuro-Symbolic Language Model

Gyan, a neuro-symbolic language model combining transformers with symbolic reasoning for improved interpretability and compositionality.

Ax Hengyu Shi, Tianyang Han, Peizhe Wang, Zhiling Wang, Xu Yang, Junhao Su 5/7/2026

Rethinking Local Learning: A Cheaper and Faster Recipe for LLM Post-Training

Local learning approach for LLM post-training reducing computation and memory costs by limiting backward gradient propagation depth.

Ax Yin Jun Phua 5/7/2026

A Foundation Model for Zero-Shot Logical Rule Induction

Neural Rule Inducer (NRI), a pretrained foundation model for zero-shot inductive logic programming enabling rule induction across tasks without retraining.

Ax Koen van der Blom, Diederick Vermetten 5/7/2026

On the Influence of the Feature Computation Budget on Per-Instance Algorithm Selection for Black-Box Optimization

Study on per-instance algorithm selection for black-box optimization, examining trade-offs between feature computation and algorithm performance.

Ax Hwiwon Lee, Jongseong Kim, Lingming Zhang 5/7/2026

Agentic Vulnerability Reasoning on Windows COM Binaries

SLYP agentic pipeline discovers race condition vulnerabilities in Windows COM binaries and generates verified proof-of-concept code.

Ax Sajal Dash, Feiyi Wang 5/7/2026

Piper: Efficient Large-Scale MoE Training via Resource Modeling and Pipelined Hybrid Parallelism

Piper system for efficient MoE model training via mathematical modeling and pipelined hybrid parallelism addressing memory and communication challenges.

Ax Srikar Kashyap Pulipaka 5/7/2026

PSK at SemEval-2026 Task 9: Multilingual Polarization Detection Using Ensemble Gemma Models with Synthetic Data Augmentation

Fine-tunes Gemma models with LoRA for multilingual polarization detection using LLM-generated synthetic data augmentation strategies.

Ax Hongyuan Zhang, Yanchen Xu, Sida Huang, Xuelong Li 5/7/2026