Isolater - Feed

Ax Chuancheng Shi, Wenhua Wu, Fei Shen, Xiaogang Zhu, Kun Hu, Zhiyong Wang 3/13/2026

OrthoEraser: Coupled-Neuron Orthogonal Projection for Concept Erasure

OrthoEraser: concept erasure method for text-to-image models using orthogonal projection to suppress unsafe concepts without collateral damage.

Ax Qizhi Chen, Chao Qi, Yihong Huang, Muquan Li, Rongzheng Wang, Dongyang Zhang, Ke Qin, Shuang Liang 3/13/2026

KEPo: Knowledge Evolution Poison on Graph-based Retrieval-Augmented Generation

KEPo: poison attack on graph-based retrieval-augmented generation systems targeting knowledge graphs to manipulate LLM outputs.

Ax Md Jahidul Islam 3/13/2026

ReHARK: Refined Hybrid Adaptive RBF Kernels for Robust One-Shot Vision-Language Adaptation

ReHARK: method for efficient one-shot adaptation of vision-language models using hybrid RBF kernels.

Ax Mayank Saini Arit Kumar Bishwas 3/13/2026

One Supervisor, Many Modalities: Adaptive Tool Orchestration for Autonomous Queries

Agentic framework for multimodal query processing with central supervisor coordinating specialized tools across text, image, audio, video, documents.

Ax Lirong Che, Shuo Wen, Shan Huang, Chuang Wang, Yuzhe Yang, Gregory Dudek, Xueqian Wang, Jian Su 3/13/2026

MANSION: Multi-floor lANguage-to-3D Scene generatIOn for loNg-horizon tasks

MANSION: language-driven framework generating multi-floor 3D environments for long-horizon robotic tasks with spatial reasoning.

Ax Ruiying Li, Yunlang Zhou, YuYao Zhu, Kylin Chen, Jingyuan Wang, Sukai Wang, Kongtao Hu, Minhui Yu, Bowen Jiang, Zhan Su, Jiayao Ma, Xin He, Yongjian Shen, Yangyang, Guanghui Ren, Maoqing Yao, Wenhao Wang, Yao Mu 3/13/2026

RoboClaw: An Agentic Framework for Scalable Long-Horizon Robotic Tasks

RoboClaw: agentic framework for long-horizon robotic tasks unifying vision-language-action systems with data collection and policy learning.

Ax Stefano Grassi 3/13/2026

How Intelligence Emerges: A Minimal Theory of Dynamic Adaptive Coordination

Dynamical theory of adaptive coordination in multi-agent systems using recursive feedback architecture.

Ax Ofir Marom 3/13/2026

UtilityMax Prompting: A Formal Framework for Multi-Objective Large Language Model Optimization

UtilityMax Prompting: formal mathematical framework for multi-objective LLM optimization using influence diagrams.

Ax Masataka Kawai, Singo Sakashita, Shumpei Ishikawa, Shogo Watanabe, Anna Matsuoka, Mikio Sakurai, Yasuto Fujimoto, Yoshiyuki Takahara, Atsushi Ohara, Hirohiko Miyake, Genichiro Ishii 3/13/2026

Performance Evaluation of Open-Source Large Language Models for Assisting Pathology Report Writing in Japanese

Evaluation of seven open-source LLMs for Japanese pathology report writing: text generation, error correction, and clinical usability.

Ax Xinhao Deng, Yixiang Zhang, Jiaqing Wu, Jiaqi Bai, Sibo Yi, Zhuoheng Zou, Yue Xiao, Rennai Qiu, Jianan Ma, Jialuo Chen, Xiaohu Du, Xiaofang Yang, Shiwen Cui, Changhua Meng, Weiqiang Wang, Jiaxing Song, Ke Xu, Qi Li 3/13/2026

Taming OpenClaw: Security Analysis and Mitigation of Autonomous LLM Agent Threats

Security analysis of autonomous LLM agents (OpenClaw), identifying attack surfaces across five-layer lifecycle and proposing mitigations.

Ax Shengyuan Liu, Zanting Ye, Yunrui Lin, Chen Hu, Wanting Geng, Xu Han, Bulat Ibragimov, Yefeng Zheng, Yixuan Yuan 3/13/2026

MedPruner: Training-Free Hierarchical Token Pruning for Efficient 3D Medical Image Understanding in Vision-Language Models

Training-free token pruning method for efficient 3D medical image processing in vision-language models.

Ax Sizhong Qin, Ramon Elias Weber, Xinzheng Lu 3/13/2026

Tokenization Allows Multimodal Large Language Models to Understand, Generate and Edit Architectural Floor Plans

HouseMind: multimodal LLM for understanding, generating, and editing architectural floor plans through tokenization approach.

Ax Gaole He, Brian Y. Lim 3/13/2026

From Control to Foresight: Simulation as a New Paradigm for Human-Agent Collaboration

Proposes simulation-based paradigm for human-agent collaboration with LLMs, enabling users to preview long-term consequences of agent actions before execution.

Ax Aleksei Petrenko, Ben Lipkin, Kevin Chen, Erik Wijmans, Marco Cusumano-Towner, Raja Giryes, Philipp Kr\"ahenb\"uhl 3/13/2026

Entropy-Preserving Reinforcement Learning

Policy gradient algorithm improvement preserving entropy and trajectory diversity during language model reasoning training.

Ax Mikel Zubillaga, Naiara Perez, Oscar Sainz, German Rigau 3/13/2026

SemBench: A Universal Semantic Framework for LLM Evaluation

Universal semantic evaluation framework for assessing LLM understanding beyond traditional perceptual metrics.

Ax Xianjing Han, Bin Zhu, Shiqi Hu, Franklin Mingzhe Li, Patrick Carrington, Roger Zimmermann, Jingjing Chen 3/13/2026

OSCBench: Benchmarking Object State Change in Text-to-Video Generation

Benchmark for evaluating text-to-video models' ability to capture object state changes specified in prompts.

Ax Konstantin Krestnikov 3/13/2026

Compression Favors Consistency, Not Truth: When and Why Language Models Prefer Correct Information

Compression-Consistency Principle explains why language models prefer correct statements: next-token prediction favors shorter internally consistent data descriptions.

Ax Erfan Mirzaei, Seyed Pooya Shariatpanahi, Alireza Tavakoli, Reshad Hosseini, Majid Nili Ahmadabadi 3/13/2026

Exploiting Expertise of Non-Expert and Diverse Agents in Social Bandit Learning: A Free Energy Approach

Studies social bandit learning where reinforcement learning agents leverage observing diverse non-expert agents using free energy optimization.

Ax Yongzhong Wang, Keyu Zhu, Yong Zhong, Liqiong Wang, Jinyu Yang, Feng Zheng 3/13/2026

RADAR: Closed-Loop Robotic Data Generation via Semantic Planning and Autonomous Causal Environment Reset

RADAR autonomous closed-loop system for robotic data generation using semantic planning and causal environment reset without human intervention.

Ax Isuri Perera, Frits de Nijs, Julian Garcia 3/13/2026

Hybrid Human-Agent Social Dilemmas in Energy Markets

Studies emergence of cooperative behavior in hybrid human-agent populations through energy load management game-theoretic scenario.

Ax Aleksander Jarz\k{e}bowicz, Adam Przyby{\l}ek, Jacinto Estima, Yen Ying Ng, Jakub Swacha, Beata Zielosko, Lech Madeyski, Noel Carroll, Kai-Kristian Kemell, Bartosz Marcinkowski, Alberto Rodrigues da Silva, Viktoria Stray, Netta Iivari, Anh Nguyen-Duc, Jorge Melegati, Boris Deliba\v{s}i\'c, Emilio Insfran 3/13/2026

The Landscape of Generative AI in Information Systems: A Synthesis of Secondary Reviews and Research Agendas

Systematic review synthesizing 28 secondary studies on generative AI adoption in organizations, identifying technical and organizational challenges.

Ax Ching-Yu Kao, Xinfeng Li, Shenyu Dai, Tianze Qiu, Pengcheng Zhou, Eric Hanchen Jiang, Philip Sperl 3/13/2026

You Told Me to Do It: Measuring Instructional Text-induced Private Data Leakage in LLM Agents

Identifies Trusted Executor Dilemma: high-privilege LLM agents executing external instructions leak private data through instruction-following vulnerability.

Ax Omar Coser 3/13/2026

ELISA: An Interpretable Hybrid Generative AI Agent for Expression-Grounded Discovery in Single-Cell Genomics

ELISA integrates scGPT expression embeddings with BioBERT for interpretable AI agent discovery in single-cell genomics data.

Ax J Alex Corll 3/13/2026

The Mirror Design Pattern: Strict Data Geometry over Model Scale for Prompt Injection Detection

Mirror design pattern for prompt injection detection using data-curation geometry for fast, deterministic, non-promptable screening.

Ax Remigiusz Kinas, Pawe{\l} Kiszczak, Sergio P. Perez, Krzysztof Ociepa, {\L}ukasz Flis, Krzysztof Wr\'obel, Adrian Gwo\'zdziej 3/13/2026

Bielik-Minitron-7B: Compressing Large Language Models via Structured Pruning and Knowledge Distillation for the Polish Language

Bielik-Minitron-7B compresses Bielik-11B model 33.4% via structured pruning and knowledge distillation for European languages.

Ax Lu Wang (The Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences, Beijing, China), Zhuoran Jin (The Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences, Beijing, China), Yupu Hao (The Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences, Beijing, China), Yubo Chen (The Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences, Beijing, China), Kang Liu (The Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences, Beijing, China), Yulong Ao (Beijing Academy of Artificial Intelligence), Jun Zhao (The Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences, Beijing, China) 3/13/2026

Think While Watching: Online Streaming Segment-Level Memory for Multi-Turn Video Reasoning in Multimodal Large Language Models

Online streaming segment-level memory system for multi-turn video reasoning in multimodal LLMs enabling concurrent perception and generation.

Ax Rajdeep Pathak, Rahul Goswami, Madhurima Panja, Palash Ghosh, Tanujit Chakraborty 3/13/2026

EnTransformer: A Deep Generative Transformer for Multivariate Probabilistic Forecasting

EnTransformer deep generative model for multivariate probabilistic time series forecasting with uncertainty quantification using transformers.

Ax Junjie Chu, Yiting Qu, Ye Leng, Michael Backes, Yun Shen, Savvas Zannettou, Yang Zhang 3/13/2026

Understanding LLM Behavior When Encountering User-Supplied Harmful Content in Harmless Tasks

Studies whether aligned LLMs refuse to proceed when encountering harmful content within benign tasks, examining content-level ethical behavior.

Ax Xingze Zou, Jing Wang, Yuhua Zheng, Xueyi Chen, Haolei Bai, Lingcheng Kong, Syed A. R. Abu-Bakar, Zhaode Wang, Chengfei Lv, Haoji Hu, Huan Wang 3/13/2026

MobileKernelBench: Can LLMs Write Efficient Kernels for Mobile Devices?

MobileKernelBench evaluates whether LLMs can generate efficient compute kernels for mobile devices, introducing benchmark dataset for systematic investigation.

Ax Zikang Ding, Haomiao Yang, Meng Hao, Wenbo Jiang, Kunlan Xiang, Runmeng Du, Yijing Liu, Ruichen Zhang, Dusit Niyato 3/13/2026

Delayed Backdoor Attacks: Exploring the Temporal Dimension as a New Attack Surface in Pre-Trained Models

Introduces delayed backdoor attacks against pre-trained models where malicious behavior activation is temporally decoupled from trigger exposure.

Ax Jiayue Pu, Zhongxiang Sun, Zilu Zhang, Xiao Zhang, Jun Xu 3/13/2026

HomeSafe-Bench: Evaluating Vision-Language Models on Unsafe Action Detection for Embodied Agents in Household Scenarios

HomeSafe-Bench: evaluation benchmark for vision-language models detecting unsafe actions in household robot scenarios.

Ax Ilias Aarab 3/13/2026

BTZSC: A Benchmark for Zero-Shot Text Classification Across Cross-Encoders, Embedding Models, Rerankers and LLMs

BTZSC benchmark comparing zero-shot text classification methods across cross-encoders, embeddings, rerankers, and instruction-tuned LLMs.

Ax Maximilian Schmidt, Swantje Plambeck, Markus Knitt, Hendrik Rose, Goerschwin Fey, Jan Christian Wieck, Stephan Balduin 3/13/2026

Flowcean - Model Learning for Cyber-Physical Systems

Flowcean: automated data-driven model generation framework for cyber-physical systems focusing on modularity and usability.

Ax Thennal D K, Chris Biemann, Hans Ole Hatzel 3/13/2026

Just Use XML: Revisiting Joint Translation and Label Projection

Framework for joint machine translation and cross-lingual label projection using XML tags; addresses degraded translation quality in prior combined approaches.

Ax Sarbartha Banerjee, Prateek Sahu, Anjo Vahldiek-Oberwagner, Jose Sanchez Vicarte, Mohit Tiwari 3/13/2026

Cascade: Composing Software-Hardware Attack Gadgets for Adversarial Threat Amplification in Compound AI Systems

Security analysis of compound AI systems combining LLMs, software tools, and databases; identifies vulnerabilities from traditional software stack and hardware layers.

Ax Xingyu Xie, Zhaochen Yu, Yue Liao, Tao Wang, Kim-Chuan Toh, Shuicheng Yan 3/13/2026

Slow-Fast Inference: Training-Free Inference Acceleration via Within-Sentence Support Stability

Slow-Fast Inference: training-free decoding acceleration leveraging stable attention support patterns within semantic spans during autoregressive generation.

Ax Sophie Jaffard, Ivo F. Sbalzarini 3/13/2026

Chemical Reaction Networks Learn Better than Spiking Neural Networks

Mathematical proof that chemical reaction networks without hidden layers outperform spiking neural networks on certain classification tasks.

Ax Zhaoyang Jiang, Zhizhong Fu, David McAllister, Yunsoo Kim, Honghan Wu 3/13/2026

Paper Title: LoV3D: Grounding Cognitive Prognosis Reasoning in Longitudinal 3D Brain MRI via Regional Volume Assessments

LoV3D: vision-language pipeline for cognitive prognosis from longitudinal brain MRI via regional volume assessments and grounded reasoning.

Ax Xiaojie Gu, Dmitry Ignatov, Radu Timofte 3/13/2026

Resource-Efficient Iterative LLM-Based NAS with Feedback Memory

LLM-based neural architecture search with feedback memory: closed-loop pipeline for iterative CNN design on consumer GPU without fine-tuning.

Ax Dimitri Staufer, Kirsten Morehouse, David Hartmann, Bettina Berendt 3/13/2026

Human-Centred LLM Privacy Audits: Findings and Frictions

LMP2: browser-based self-audit tool for inspecting LLM associations with individuals, with user study findings on privacy and model behavior.

Ax Taeho Lee, Donghwan Lee 3/13/2026

Taming the Adversary: Stable Minimax Deep Deterministic Policy Gradient via Fractional Objectives

Minimax deep deterministic policy gradient: RL algorithm for stable performance under external disturbances using fractional objectives.

Ax William Brach, Tomas Bedej, Jacob Nielsen, Jacob Pichna, Juraj Bedej, Eemeli Saarensilta, Julie Dupouy, Gianluca Barmina, Andrea Blasi N\'u\~nez, Peter Schneider-Kamp, Kristian Ko\v{s}\v{t}\'al, Michal Ries, Lukas Galke Poech 3/13/2026

SommBench: Assessing Sommelier Expertise of Language Models

SommBench: multilingual benchmark assessing LLM capabilities in sommelier expertise, evaluating cultural knowledge beyond linguistic encoding.

Ax Seth Karten, Rahul Dev Appapogu, Chi Jin 3/13/2026

Automatic Generation of High-Performance RL Environments

Agent-assisted code generation for translating complex RL environments into high-performance implementations with iterative repair and verification.

Ax Zhoujun Cheng, Yutao Xie, Yuxiao Qu, Amrith Setlur, Shibo Hao, Varad Pimpalkhute, Tongtong Liang, Feng Yao, Zhengzhong Liu, Eric Xing, Virginia Smith, Ruslan Salakhutdinov, Zhiting Hu, Taylor Killian, Aviral Kumar 3/13/2026

IsoCompute Playbook: Optimally Scaling Sampling Compute for LLM RL

IsoCompute Playbook: scaling laws for optimal compute allocation in LLM reinforcement learning post-training across rollouts, batches, and update steps.

Ax Zexuan Yan, Jiarui Jin, Yue Ma, Shijian Wang, Jiahui Hu, Wenxiang Jiao, Yuan Lu, Linfeng Zhang 3/13/2026

GlyphBanana: Advancing Precise Text Rendering Through Agentic Workflows

GlyphBanana: agentic workflow for precise text and formula rendering in generative models, addressing out-of-distribution instruction-following challenges.

Ax Krishnakumar Balasubramanian, Shiva Prasad Kasiviswanathan 3/13/2026

A Quantitative Characterization of Forgetting in Post-Training

Theoretical analysis of catastrophic forgetting in continual post-training of generative models, formalizing mass forgetting and component drift mechanisms.

Ax Jingyang Ke, Weihan Li, Amartya Pradhan, Jeffrey Markowitz, Anqi Wu 3/13/2026

BehaviorVLM: Unified Finetuning-Free Behavioral Understanding with Vision-Language Reasoning

BehaviorVLM: vision-language framework for animal pose estimation and behavioral understanding without human annotation using finetuning-free approach.

Ax {\L}ukasz Borchmann, Jordy Van Landeghem, Micha{\l} Turski, Shreyansh Padarha, Ryan Othniel Kearns, Adam Mahdi, Niels Rogge, Cl\'ementine Fourrier, Siwei Han, Huaxiu Yao, Artemis Llabr\'es, Yiming Xu, Dimosthenis Karatzas, Hao Zhang, Anupam Datta 3/13/2026