Isolater - Feed

Ax Jin Liu, Yinbin Miao, Ning Xi, Junkang Liu 2/24/2026

Rethinking LoRA for Privacy-Preserving Federated Learning in Large Models

Analysis of LoRA application in privacy-preserving federated learning for large vision and language models, revealing privacy-utility trade-offs.

Ax Jin Liu, Yinbin Miao, Ning Xi, Junkang Liu 2/24/2026

DP-FedAdamW: An Efficient Optimizer for Differentially Private Federated Large Models

DP-FedAdamW optimizer for federated learning of large models under differential privacy, addressing variance and convergence issues.

Ax Krzysztof Adamkiewicz, Brian Moser, Stanislav Frolov, Tobias Christian Nauen, Federico Raue, Andreas Dengel 2/24/2026

When Pretty Isn't Useful: Investigating Why Modern Text-to-Image Models Fail as Reliable Training Data Generators

Study showing text-to-image diffusion models produce visually appealing but unreliable synthetic training data for computer vision tasks.

Ax Ian Steenstra, Paola Pedrelli, Weiyan Shi, Stacy Marsella, Timothy W. Bickmore 2/24/2026

Assessing Risks of Large Language Models in Mental Health Support: A Framework for Automated Clinical AI Red Teaming

Red teaming framework for evaluating risks of LLMs in mental health support using AI psychotherapists and simulated patient agents.

Ax Moritz A. Zanger, Yijun Wu, Pascal R. Van der Vaart, Wendelin B\"ohmer, Matthijs T. J. Spaan 2/24/2026

On the Equivalence of Random Network Distillation, Deep Ensembles, and Bayesian Inference

Theoretical analysis showing equivalence between random network distillation, deep ensembles, and Bayesian inference for uncertainty quantification.

Ax Yuxing Tian, Fengran Mo, Weixu Zhang, Yiyan Qi, Jian-Yun Nie 2/24/2026

ReAttn: Improving Attention-based Re-ranking via Attention Re-weighting

ReAttn method improving LLM-based re-ranking by re-weighting attention mechanisms to address concentration and shallow processing issues.

Ax Zachary Ravichadran, David Snyder, Alexander Robey, Hamed Hassani, Vijay Kumar, George J. Pappas 2/24/2026

Contextual Safety Reasoning and Grounding for Open-World Robots

Safety framework (CORE) for robots reasoning about contextual safety requirements in open-world environments with varying conditions.

Ax Nuocheng Yang, Sihua Wang, Zhaohui Yang, Mingzhe Chen, Changchuan Yin, Kaibin Huang 2/24/2026

A Secure and Private Distributed Bayesian Federated Learning Design

Federated learning framework addressing privacy, convergence, and Byzantine robustness in distributed model training without central server.

Ax Yuxing Tian, Yiyan Qi, Fengran Mo, Weixu Zhang, Jian Guo, Jian-Yun Nie 2/24/2026

Learning Discriminative and Generalizable Anomaly Detector for Dynamic Graph with Limited Supervision

Semi-supervised method for detecting anomalies in dynamic graphs with limited labeled data, balancing unsupervised and supervised approaches.

Ax Fahmida Liza Piya, Rahmatollah Beheshti 2/24/2026

AgenticSum: An Agentic Inference-Time Framework for Faithful Clinical Text Summarization

AgenticSum: agentic inference-time framework using LLMs for clinical text summarization with multi-step verification to reduce hallucinations.

Ax Apoorva Vashisth (Purdue University), Manav Kulshrestha (Purdue University), Pranav Bakshi (IIT Kharagpur), Damon Conover (DEVCOM Army Research Lab), Guillaume Sartoretti (National University of Singapore), Aniket Bera (Purdue University) 2/24/2026

To Move or Not to Move: Constraint-based Planning Enables Zero-Shot Generalization for Interactive Navigation

Robot navigation system that manipulates obstacles to find paths in cluttered environments using constraint-based planning.

Ax Ge Yuan, Qiyuan Qiao, Jing Zhang, Dong Xu 2/24/2026

AdaWorldPolicy: World-Model-Driven Diffusion Policy with Online Adaptive Learning for Robotic Manipulation

World-model-driven diffusion policy framework with online adaptive learning for robotic manipulation under dynamic conditions.

Ax Zac Garby, Andrew D. Gordon, David Sands 2/24/2026

The LLMbda Calculus: AI Agents, Conversations, and Information Flow

Formal analysis of information flow and attack surface in AI agent systems, showing how malicious prompt injection compromises multi-step conversations.

Ax Natalia Moskvina, Raquel Montero, Masaya Yoshida, Ferdy Hubers, Paolo Morosi, Walid Irhaymi, Jin Yan, Tamara Serrano, Elena Pagliarini, Fritz G\"unther, Evelina Leivada 2/24/2026

Multilingual Large Language Models do not comprehend all natural languages to equal degrees

Study benchmarking LLM comprehension across multilingual datasets, showing significant performance gaps in low-resource non-Western languages.

Ax Kundan Thota, Xuanhao Mu, Thorsten Schlachter, Veit Hagenmeyer 2/24/2026

HeatPrompt: Zero-Shot Vision-Language Modeling of Urban Heat Demand from Satellite Images

Zero-shot vision-language framework estimating urban heat demand from satellite imagery using pretrained LVLM features.

Ax Shan Yang, Yang Liu 2/24/2026

Descent-Guided Policy Gradient for Scalable Cooperative Multi-Agent Learning

Policy gradient method for multi-agent reinforcement learning reducing cross-agent noise variance for scalable cooperative learning.

Ax Thanh Q. Tran, Arun Verma, Kiwan Wong, Bryan Kian Hsiang Low, Daniela Rus, Wei Xiao 2/24/2026

BarrierSteer: LLM Safety via Learning Barrier Steering

BarrierSteer framework for LLM safety using barrier functions and steering, addressing adversarial attacks and unsafe content generation.

Ax Kairan Zhao, Iurie Luca, Peter Triantafillou 2/24/2026

Benchmarking Unlearning for Vision Transformers

Benchmark for machine unlearning methods on Vision Transformers, extending prior CNN-focused unlearning research to transformer architectures.

Ax Jiahui Fu, Junyu Nan, Lingfeng Sun, Hongyu Li, Jianing Qian, Jennifer L. Barry, Kris Kitani, George Konidaris 2/24/2026

NovaPlan: Zero-Shot Long-Horizon Manipulation via Closed-Loop Video Language Planning

NovaPlan hierarchical framework combines vision-language models with video planning and geometric robot control for long-horizon manipulation tasks.

Ax Lingwei Gu, Nour Jedidi, Jimmy Lin 2/24/2026

NanoKnow: How to Know What Your Language Model Knows

NanoKnow analyzes how LLMs encode knowledge by studying nanochat models with fully open, transparent pre-training data.

Ax Zaifu Zhan, Min Zeng, Shuang Zhou, Yiran Song, Xiaoyi Chen, Yu Hou, Yifan Wu, Yang Ruan, Rui Zhang 2/24/2026

To Reason or Not to: Selective Chain-of-Thought in Medical Question Answering

Selective Chain-of-Thought improves medical QA efficiency by predicting whether questions require reasoning before generating rationales.

Ax Mert Cemri, Shubham Agrawal, Akshat Gupta, Shu Liu, Audrey Cheng, Qiuyang Mang, Ashwin Naren, Lutfi Eren Erdogan, Koushik Sen, Matei Zaharia, Alex Dimakis, Ion Stoica 2/24/2026

AdaEvolve: Adaptive LLM Driven Zeroth-Order Optimization

AdaEvolve uses LLMs as semantic mutation operators in evolutionary optimization with adaptive scheduling for automated program generation.

Ax Mohammad Amanlou, Erfan Shafiee Moghaddam, Yasaman Amou Jafari, Mahdi Noori, Farhan Farsi, Behnam Bahrak 2/24/2026

KNIGHT: Knowledge Graph-Driven Multiple-Choice Question Generation with Adaptive Hardness Calibration

KNIGHT framework generates multiple-choice question datasets from knowledge graphs and external sources using LLMs for RAG system evaluation.

Ax Zehao Wang, Mingzhe Han, Wei Cheng, Yue-Kai Huang, Philip Ji, Denton Wu, Mahdi Safari, Flemming Holtorf, Kenaish AlQubaisi, Norbert M. Linke, Danyang Zhuo, Yiran Chen, Ting Wang, Dirk Englund, Tingjun Chen 2/24/2026

Agentic AI for Scalable and Robust Optical Systems Control

AgentOptics framework enables agentic AI for autonomous optical system control using Model Context Protocol with 64 MCP tools and 410-task benchmark.

Ax Zhenyao Ma, Yue Liang, Dongxu Li 2/24/2026

Behavior Learning (BL): Learning Hierarchical Optimization Structures from Data

Behavior Learning framework learns interpretable hierarchical optimization structures from data with identifiability guarantees.

Ax Maijunxian Wang, Ruisi Wang, Juyi Lin, Ran Ji, Thadd\"aus Wiedemer, Qingying Gao, Dezhi Luo, Yaoyao Qian, Lianyu Huang, Zelong Hong, Jiahui Ge, Qianli Ma, Hang He, Yifan Zhou, Lingzi Guo, Lantao Mei, Jiachen Li, Hanwen Xing, Tianqi Zhao, Fengyuan Yu, Weihang Xiao, Yizheng Jiao, Jianheng Hou, Danyang Zhang, Pengcheng Xu, Boyang Zhong, Zehong Zhao, Gaoyun Fang, John Kitaoka, Yile Xu, Hua Xu, Kenton Blacutt, Tin Nguyen, Siyuan Song, Haoran Sun, Shaoyue Wen, Linyang He, Runming Wang, Yanzhi Wang, Mengyue Yang, Ziqiao Ma, Rapha\"el Milli\`ere, Freda Shi, Nuno Vasconcelos, Daniel Khashabi, Alan Yuille, Yilun Du, Ziming Liu, Bo Li, Dahua Lin, Ziwei Liu, Vikash Kumar, Yijiang Li, Lei Yang, Zhongang Cai, Hokin Deng 2/24/2026

A Very Big Video Reasoning Suite

Benchmark suite evaluating video model reasoning capabilities on spatiotemporal understanding including continuity, interaction, and causality.

Ax Henry Hengyuan Zhao, Kaiming Yang, Wendi Yu, Difei Gao, Mike Zheng Shou 2/24/2026

WorldGUI: An Interactive Benchmark for Desktop GUI Automation from Any Starting Point

WorldGUI benchmark evaluates GUI agents on desktop automation tasks from arbitrary starting points with partial configurations and state variability.

Ax Sanidhya Vijayvargiya, Xuhui Zhou, Akhila Yerukola, Maarten Sap, Graham Neubig 2/24/2026

Ambig-SWE: Interactive Agents to Overcome Underspecificity in Software Engineering

Ambig-SWE studies LLM agent capabilities to handle underspecified instructions by asking clarifying questions in interactive software engineering tasks.

Ax Gaole Dai, Shiqi Jiang, Ting Cao, Yuanchun Li, Yuqing Yang, Rui Tan, Mo Li, Lili Qiu 2/24/2026

Advancing Mobile GUI Agents: A Verifier-Driven Approach to Practical Deployment

V-Droid is a mobile GUI automation agent using LLMs as verifiers to evaluate candidate actions before execution, with discretized action space framework.

Ax Seungyoon Woo, Junhyeog Yun, Gunhee Kim 2/24/2026

Meta-Continual Learning of Neural Fields

Novel meta-continual learning approach for neural fields using modular architecture and optimization-based meta-learning to address catastrophic forgetting.

Ax Zuhong Lin, Daoyuan Ren, Kai Ran, Jing Sun, Songlin Yu, Xuefeng Bai, Xiaotian Huang, Haiyang He, Pengxu Pan, Ying Fang, Zhanglin Li, Haipu Li, Jingjing Yao 2/24/2026

Reshaping MOFs text mining with a dynamic multi-agents framework of large language model

MOFh6 uses LLMs to extract and standardize metal-organic framework synthesis conditions from raw articles and crystal codes into structured tables.

Ax Georgy Noarov, Soham Mallick, Tao Wang, Sunay Joshi, Yan Sun, Yangxinyu Xie, Mengxin Yu, Edgar Dobriban 2/24/2026

Foundations of Top-$k$ Decoding For Language Models

Theoretical analysis of top-k decoding method for LLM sampling, validating sparsity assumptions in next-token distributions.

Ax Florian Andreas Marwitz, Tanya Braun, Ralf M\"oller, Marcel Gehrke 2/24/2026

Lifted Forward Planning in Relational Factored Markov Decision Processes with Concurrent Actions

Relational first-order representation (Foreplan) for efficient forward planning in factored MDPs with concurrent actions and objects.

Ax Subhrangshu Nandi, Arghya Datta, Rohith Nama, Udita Patel, Nikhil Vichare, Indranil Bhattacharya, Prince Grover, Shivam Asija, Giuseppe Carenini, Wei Zhang, Arushi Gupta, Sreyoshi Bhaduri, Jing Xu, Huzefa Raja, Shayan Ray, Aaron Chan, Esther Xu Fei, Gaoyuan Du, Zuhaib Akhtar, Harshita Asnani, Weian Chan, Ming Xiong, Francesco Carbone, Jeetu Mirchandani 2/24/2026

SOP-Bench: Complex Industrial SOPs for Evaluating LLM Agents

SOP-Bench: 2000+ task benchmark for evaluating LLM agents on complex multi-step industrial procedures across 12 business domains.

Ax Ziyang Guo, Berk Ustun, Jessica Hullman 2/24/2026

Explanations are a Means to an End: Decision Theoretic Explanation Evaluation

Decision-theoretic framework for evaluating model explanations based on actual decision task performance improvement.

Ax Kaizhen Tan, Yufan Wu, Yuxuan Liu, Haoran Zeng 2/24/2026

Decoding Tourist Perception in Historic Urban Quarters with Multimodal Social Media Data: An AI-Based Framework and Evidence from Shanghai

AI framework using multimodal social media data to analyze tourist perception and preferences in historic urban areas.

Ax Donghwan Lee, Hyukjun Yang, Bum Geun Park 2/24/2026