Isolater - Feed

Ax Zhongcan Xiao (Neutron Scattering Division, Oak Ridge National Laboratory, Oak Ridge, Tennesse USA), Leyi Zhang (Neutron Scattering Division, Oak Ridge National Laboratory, Oak Ridge, Tennesse USA, Department of Linguistics, University of Illinois Urbana-Champaign, Urbana, Illinois, USA), Guannan Zhang (Computer Science and Mathematics Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee, USA), Xiaoping Wang (Neutron Scattering Division, Oak Ridge National Laboratory, Oak Ridge, Tennesse USA) 2/20/2026

NeuDiff Agent: A Governed AI Workflow for Single-Crystal Neutron Crystallography

NeuDiff Agent LLM-based workflow for automated analysis and reporting in neutron crystallography at Spallation Neutron Source.

Ax Eiman Kanjo, Mustafa Aslanov 2/20/2026

Node Learning: A Framework for Adaptive, Decentralised and Collaborative Network Edge AI

Node Learning decentralized paradigm for edge AI where intelligence resides at individual nodes without centralized servers.

Ax Priyaranjan Pattnayak, Sanchari Chowdhuri 2/20/2026

IndicJR: A Judge-Free Benchmark of Jailbreak Robustness in South Asian Languages

IndicJR judge-free benchmark of jailbreak robustness across 12 Indic/South Asian languages covering 45,216 adversarial prompts.

Ax Haiyang Xu, Xi Zhang, Haowei Liu, Junyang Wang, Zhaozai Zhu, Shengjie Zhou, Xuhao Hu, Feiyu Gao, Junjie Cao, Zihua Wang, Zhiyuan Chen, Jitong Liao, Qi Zheng, Jiahui Zeng, Ze Xu, Shuai Bai, Junyang Lin, Jingren Zhou, Ming Yan 2/20/2026

Mobile-Agent-v3.5: Multi-platform Fundamental GUI Agents

GUI-Owl-1.5 native GUI agent model in multiple sizes supporting desktop, mobile, browser with state-of-the-art results on 20+ automation benchmarks.

Ax Hongwei Li, Zhun Wang, Qinrun Dai, Yuzhou Nie, Jinjun Peng, Ruitong Liu, Jingyang Zhang, Kaijie Zhu, Jingxuan He, Lun Wang, Yangruibo Ding, Yueqi Chen, Wenbo Guo, Dawn Song 2/20/2026

OpenSage: Self-programming Agent Generation Engine

OpenSage first agent development kit with self-programming capability for automatically designing agent topology, tools, and memory components.

Ax Tanqiu Jiang, Yuhui Wang, Jiacheng Liang, Ting Wang 2/20/2026

AgentLAB: Benchmarking LLM Agents against Long-Horizon Attacks

AgentLAB benchmark for evaluating LLM agent vulnerabilities to adaptive long-horizon attacks in complex multi-turn environments.

Ax Juliusz Ziomek, William Bankes, Lorenz Wolf, Shyam Sundhar Ramesh, Xiaohang Tang, Ilija Bogunovic 2/20/2026

LLM-WikiRace: Benchmarking Long-term Planning and Reasoning over Real-World Knowledge Graphs

LLM-WikiRace benchmark evaluates planning, reasoning, and world knowledge by requiring models to navigate Wikipedia hyperlinks from source to target page.

Ax Idhant Gulati, Shivam Raval 2/20/2026

Narrow fine-tuning erodes safety alignment in vision-language agents

Study showing fine-tuning vision-language agents on narrow tasks causes emergent misalignment that generalizes across unrelated domains and modalities.

Ax Justin Albrethsen, Yash Datta, Kunal Kumar, Sharath Rajasekar 2/20/2026

DeepContext: Stateful Real-Time Detection of Multi-Turn Adversarial Intent Drift in LLMs

DeepContext stateful monitoring framework for detecting adversarial intent drift across multi-turn LLM dialogues, addressing safety gaps in sequential interactions.

Ax Hexi Jin, Stephen Liu, Yuheng Li, Simran Malik, Yiying Zhang 2/20/2026

SourceBench: Can AI Answers Reference Quality Web Sources?

SourceBench evaluates quality of web sources cited by LLMs across 100 queries using eight-metric framework beyond correctness.

Ax Arnold Cartagena, Ariane Teixeira 2/20/2026

Mind the GAP: Text Safety Does Not Transfer to Tool-Call Safety in LLM Agents

GAP benchmark reveals that text-level safety alignment in LLM agents doesn't transfer to tool-call safety, measuring real-world action harms.

Ax Hejia Zhang, Zhongming Yu, Chia-Tung Ho, Haoxing Ren, Brucek Khailany, Jishen Zhao 2/20/2026

LLM4Cov: Execution-Aware Agentic Learning for High-coverage Testbench Generation

LLM4Cov framework for offline agent learning applied to high-coverage hardware testbench generation using non-differentiable execution feedback.

Ax Xinhao Deng, Jiaqing Wu, Miao Chen, Yue Xiao, Ke Xu, Qi Li 2/20/2026

Automating Agent Hijacking via Structural Template Injection

Phantom: automated agent hijacking attack on LLM agents via structural template injection, addressing OWASP-highlighted threat with improved transferability.

Ax Vishal Srivastava 2/20/2026

Fundamental Limits of Black-Box Safety Evaluation: Information-Theoretic and Computational Barriers from Latent Context Conditioning

Theoretical analysis of fundamental limits in black-box safety evaluation of AI systems, showing latent context-conditioned policies create evaluation gaps.

Ax Yan Wang, Yi Han, Lingfei Qian, Yueru He, Xueqing Peng, Dongji Feng, Zhuohan Xie, Vincent Jim Zhang, Rosie Guo, Fengran Mo, Jimin Huang, Yankai Chen, Xue Liu, Jian-Yun Nie 2/20/2026

Conv-FinRe: A Conversational and Longitudinal Benchmark for Utility-Grounded Financial Recommendation

Conv-FinRe benchmark for stock recommendation that evaluates utility-grounded decisions rather than behavioral imitation in conversational finance advisory.

Ax Zhao Tan, Yiji Zhao, Shiyu Wang, Chang Xu, Yuxuan Liang, Xiping Liu, Shirui Pan, Ming Jin 2/20/2026

Sonar-TS: Search-Then-Verify Natural Language Querying for Time Series Databases

Sonar-TS neuro-symbolic framework for natural language querying time series databases, handling morphological intents and ultra-long histories.

Ax Zichen Wang, Wanli Ma, Zhenyu Ming, Gong Zhang, Kun Yuan, Zaiwen Wen 2/20/2026

M2F: Automated Formalization of Mathematical Literature at Scale

M2F agentic framework for end-to-end project-scale autoformalization of mathematics in Lean, managing cross-file dependencies and imports.

Ax Deepanjan Bhol 2/20/2026

Sales Research Agent and Sales Research Bench

AI agent for Microsoft Dynamics 365 Sales querying live CRM data, reasoning over schemas, and producing decision-ready insights with benchmarking.

Ax Shengtian Yang (Southeast University, Kuaishou Technology), Yu Li (Southeast University), Shuo He (Nanyang Technological University), Yewen Li (Kuaishou Technology), Qingpeng Cai (Kuaishou Technology), Peng Jiang (Kuaishou Technology), Lei Feng (Southeast University) 2/20/2026

Phase-Aware Mixture of Experts for Agentic Reinforcement Learning

Mixture-of-Experts architecture for RL policy networks in LLM agents, addressing simplicity bias by allocating capacity across task complexity.

Ax Uria Franko 2/20/2026

Dynamic System Instructions and Tool Exposure for Efficient Agentic LLMs

Instruction-Tool Retrieval (ITR) RAG variant dynamically retrieves minimal system prompts and necessary tool subsets per step for efficient agentic LLMs.

Ax Seoyoung Lee, Seobin Yoon, Seongbeen Lee, Yoojung Chun, Dayoung Park, Doyeon Kim, Joo Yong Sim 2/20/2026

IntentCUA: Learning Intent-level Representations for Skill Abstraction and Multi-Agent Planning in Computer-Use Agents

Multi-agent computer-use framework with intent-aligned plan memory to stabilize long-horizon execution and reduce error accumulation.

Ax Yunseok Han, Yejoon Lee, Jaeyoung Do 2/20/2026

RFEval: Benchmarking Reasoning Faithfulness under Counterfactual Reasoning Intervention in Large Reasoning Models

Framework evaluating reasoning faithfulness in large reasoning models through counterfactual intervention on stance consistency and causal influence.

Ax Yonghyeon Jo, Sunwoo Lee, Seungyul Han 2/20/2026

Retaining Suboptimal Actions to Follow Shifting Optima in Multi-Agent Reinforcement Learning

Multi-agent RL method retaining multiple high-value actions via sub-value functions to adapt to shifting value functions.

Ax Sumedh Rasal 2/20/2026

Predictive Batch Scheduling: Accelerating Language Model Training Through Loss-Aware Sample Prioritization

Predictive Batch Scheduling uses lightweight online predictor to prioritize high-loss samples, accelerating LLM training convergence.

Ax Kan Watanabe, Rikuto Tsuchida, Takahiro Monno, Bin Huang, Kazuma Yamasaki, Youmei Fan, Kazumasa Shimari, Kenichi Matsumoto 2/20/2026

How AI Coding Agents Communicate: A Study of Pull Request Description Characteristics and Human Review Responses

Empirical study analyzing pull request characteristics from five AI coding agents and human reviewer responses using AIDev dataset.

Ax Zhaoyang Li, Xingzhi Jin, Junyu Pan, Qianqian Yang, Zhiguo Shi 2/20/2026

Agentic Wireless Communication for 6G: Intent-Aware and Continuously Evolving Physical-Layer Intelligence

6G wireless systems architecture using intent-driven autonomous agents for multi-dimensional objectives and evolving requirements.

Ax Xiangyu Zhou, Chenhan Xiao, Yang Weng 2/20/2026

Owen-based Semantics and Hierarchy-Aware Explanation (O-Shap)

Owen-value based method extending SHAP for hierarchical feature attribution in vision tasks with spatial/semantic dependencies.

Ax Abdulrahman AlRabah, Priyanka Kargupta, Jiawei Han, Abdussalam Alawini 2/20/2026

Instructor-Aligned Knowledge Graphs for Personalized Learning

Knowledge graphs capturing educational concept dependencies and prerequisites for personalized learning at scale.

Ax Ilya Levin 2/20/2026

Epistemology of Generative AI: The Geometry of Knowing

Philosophical examination of generative AI's epistemic character and implications for knowledge production in science, education, and institutions.

Ax Sicheng Mao 2/20/2026

Texo: Formula Recognition within 20M Parameters

Texo: minimalist 20M parameter formula recognition model achieving state-of-the-art performance with 80% size reduction through distillation and transfer learning.

Ax Enrique Crespo-Fernandez, Oliver Ray, Telmo de Menezes e Silva Filho, Peter Flach 2/20/2026

Continual learning and refinement of causal models through dynamic predicate invention

Framework for constructing symbolic causal world models online by integrating continuous model learning with meta-interpretive learning in agent decision loops.

Ax Yi-Chih Huang 2/20/2026

From Labor to Collaboration: A Methodological Experiment Using AI Agents to Augment Research Perspectives in Taiwan's Humanities and Social Sciences

Methodological experiment using AI agents in collaborative research workflows for humanities and social sciences, analyzing Taiwan Claude.ai usage data.

Ax Ben Yellin, Ehud Ezra, Mark Foreman, Shula Grinapol 2/20/2026

Decoding the Human Factor: High Fidelity Behavioral Prediction for Strategic Foresight

Framework for predicting consistent individual-specific human behavior in high-stakes environments by combining LLMs with psychological trait modeling.

Ax Bianca Raimondi, Maurizio Gabbrielli 2/20/2026

Mechanistic Interpretability of Cognitive Complexity in LLMs via Linear Probing using Bloom's Taxonomy

Mechanistic interpretability study using linear probing and Bloom's Taxonomy to analyze cognitive complexity in LLM internal neural representations.

Ax Zeyu Zhang, Ryan Chen, Bradly C. Stadie 2/20/2026

All Leaks Count, Some Count More: Interpretable Temporal Contamination Detection in LLM Backtesting

Framework for detecting and quantifying temporal data contamination in LLM backtesting to validate whether models leak post-cutoff training knowledge.

Ax Linxi Jiang, Rui Xi, Zhijie Liu, Shuo Chen, Zhiqiang Lin, Suman Nath 2/20/2026

Web Verbs: Typed Abstractions for Reliable Task Composition on the Agentic Web

Web Verbs framework providing typed abstractions for reliable task composition on agentic web, enabling LLM-based web agents beyond low-level primitives.

Ax Anuj Gupta 2/20/2026

ArXiv-to-Model: A Practical Study of Scientific LM Training

Case study training 1.36B-parameter scientific language model from raw arXiv LaTeX sources, documenting end-to-end process for domain-specialized LM development.

Ax Hui Min Wong, Philip Heesen, Pascal Janetzky, Martin Bendszus, Stefan Feuerriegel 2/20/2026

MedClarify: An information-seeking AI agent for medical diagnosis with case-specific follow-up questions

MedClarify: LLM-based AI agent for medical diagnosis that iteratively asks follow-up questions to resolve diagnostic uncertainty through differential reasoning.

Ax Angelo Porrello, Pietro Buzzega, Felix Dangel, Thomas Sommariva, Riccardo Salami, Lorenzo Bonicelli, Simone Calderara 2/20/2026

Dataless Weight Disentanglement in Task Arithmetic via Kronecker-Factored Approximate Curvature

Method for disentangling task vectors in foundation models using Kronecker-factored approximate curvature without external task data.

Ax Adri\`a Molina, Oriol Ramos Terrades, Josep Llad\'os 2/20/2026

Visual Model Checking: Graph-Based Inference of Visual Routines for Image Retrieval

Graph-based visual inference approach for complex image retrieval queries involving relationships, compositions, and precise constraints.

Ax Diana Addae, Diana Rogachova, Nafiseh Kahani, Masoud Barati, Michael Christensen, Chen Zhou 2/20/2026

A Privacy by Design Framework for Large Language Model-Based Applications for Children

Privacy-by-Design framework for LLM-based applications targeting children, addressing implementation gaps in privacy regulation compliance.

Ax Pranay Jain, Maximilian Kasper, G\"oran K\"ober, Axel Plinge, Dominik Seu{\ss} 2/20/2026

Pareto Optimal Benchmarking of AI Models on ARM Cortex Processors for Sustainable Embedded Systems

Benchmarking framework for optimizing AI models on ARM Cortex embedded processors, measuring energy efficiency, accuracy, and resource utilization.

Ax Dun Yuan (Charlie), Hao Zhou (Charlie), Xue Liu (Charlie), Hao Chen (Charlie), Yan Xin (Charlie), Jianzhong (Charlie), Zhang 2/20/2026

Enhancing Large Language Models (LLMs) for Telecom using Dynamic Knowledge Graphs and Explainable Retrieval-Augmented Generation

arXiv paper on applying LLMs to telecom domain using dynamic knowledge graphs and retrieval-augmented generation to reduce hallucinations and improve accuracy.

Ax Shashank Aggarwal, Ram Vikas Mishra, Amit Awekar 2/20/2026

Evaluating Chain-of-Thought Reasoning through Reusability and Verifiability

Evaluation framework for chain-of-thought reasoning quality using reusability and verifiability metrics in multi-agent IR pipelines.

Ax Yue Liu, Zhiyuan Hu, Flood Sung, Jiaheng Zhang, Bryan Hooi 2/20/2026

KLong: Training LLM Agent for Extremely Long-horizon Tasks

KLong open-source LLM agent framework for extremely long-horizon tasks using trajectory-splitting SFT and progressive RL training.

Ax Hongjue Zhao, Haosen Sun, Jiangtao Kong, Xiaochang Li, Qineng Wang, Liwei Jiang, Qi Zhu, Tarek Abdelzaher, Yejin Choi, Manling Li, Huajie Shao 2/20/2026

ODESteer: A Unified ODE-Based Steering Framework for LLM Alignment

ODESteer unified ODE-based framework for LLM alignment via activation steering with multi-step guidance.

Ax Asif Hasan Chowdhury, Md. Fahim Islam, M Ragib Anjum Riad, Faiyaz Bin Hashem, Md Tanzim Reza, Md. Golam Rabiul Alam 2/20/2026

A Hybrid Federated Learning Based Ensemble Approach for Lung Disease Diagnosis Leveraging Fusion of SWIN Transformer and CNN

Federated learning ensemble combining SWIN Transformer and CNN for lung disease diagnosis from medical imaging.

Ax Lance Ying, Ryan Truong, Prafull Sharma, Kaiya Ivy Zhao, Nathan Cloos, Kelsey R. Allen, Thomas L. Griffiths, Katherine M. Collins, Jos\'e Hern\'andez-Orallo, Phillip Isola, Samuel J. Gershman, Joshua B. Tenenbaum 2/20/2026