The Geometry of Representational Failures in Vision Language Models
Dipartimento di Fisica, Università di Torino（都灵大学物理系）；Princeton Neuroscience Institute and AI Lab, Princeton University（普林斯顿大学神经科学研究所和AI实验室）；Intesa Sanpaolo AI Research（Intesa Sanpaolo AI研究中心）；Dipartimento di Scienze Matematiche, Politecnico di Torino（都灵理工学院数学科学系）；Network Science Institute, Northeastern University London, UK（伦敦大学东北方大学网络科学研究所）
Daniele Savietto, Declan Campbell, André Panisson, Marco Nurisso, Giovanni Petri, Jonathan D. Cohen, Alan Perotti
Vision-Language Models (VLMs) exhibit puzzling failures in multi-object visual tasks, such as hallucinating non-existent elements or failing to identify the most similar objects among distractions. While these errors mirror human cognitive constraints, such as the 'Binding Problem', the internal mechanisms driving them in artificial systems remain poorly understood. Here, we propose a mechanistic insight by analyzing the representational geometry of open-weight VLMs (Qwen, InternVL, Gemma), comparing methodologies to distill "concept vectors'' - latent directions encoding visual concepts. We validate our concept vectors via steering interventions that reliably manipulate model behavior in both simplified and naturalistic vision tasks (e.g., forcing the model to perceive a red flower as blue). We observe that the geometric overlap between these vectors strongly correlates with specific error patterns, offering a grounded quantitative framework to understand how internal representations shape model behavior and drive visual failures.
MMDrive: Interactive Scene Understanding Beyond Vision with Multi-representational Fusion
organization= College of Computer Science；Technology, Jilin University , city= Changchun , country= China；organization= Georgia Institute of Technology , city= Atlanta , country= USA；organization= Qingdao Institute of Software, College of Computer Science；Technology, China University of Petroleum (East China) , city= Qingdao , country= China；organization= Institute for Math \& AI, Wuhan University , city= Wuhan , country= China；organization= University of Michigan, Ann Arbor , country= USA；organization= Thrust of Artificial Intelligence, Hong Kong University of Science；organization= School of Artificial Intelligence；Computer Science, Nantong University , city= Nantong , country= China
Discovering Failure Modes in Vision-Language Models using RL
Mila – Québec AI Institute（魁北克AI研究所）；Université de Montréal（蒙特利尔大学）；Michigan State University（密歇根州立大学）；McGill University（麦吉尔大学）；Canada CIFAR AI Chair（加拿大CIFAR人工智能主席）
From Representational Complementarity to Dual Systems: Synergizing VLM and Vision-Only Backbones for End-to-End Driving
Department of Automation, University of Science and Technology of China（中国科学技术大学自动化系）；School of Electronic Information Engineering, Beihang University（北京航空航天大学电子信息工程学院）；School of Artificial Intelligence and Automation, Huazhong University of Science and Technology（华中科技大学人工智能与自动化学院）；National Superior College for Engineers, Beihang University（北京航空航天大学国家级工程师学院）；Beijing Academy of Artificial Intelligence（北京人工智能研究院）；Lenovo Group Limited（联想集团有限公司）；Institute for AI Industry Research, Tsinghua University（清华大学人工智能产业研究院）
Your Vision-Language Model Can't Even Count to 20: Exposing the Failures of VLMs in Compositional Counting
Guilin University of Electronic Technology（桂林电子科技大学）；The Ohio State University（俄亥俄州立大学）；University of Wisconsin-Madison（威斯康星大学麦迪逊分校）；University of California, Berkeley（加州大学伯克利分校）
Feature-Aware Token Attack for Compression-Triggered Stealthy Failures in Large Vision-Language Models
Beijing University of Posts and Telecommunications（北京邮电大学）；Jiangnan University（江南大学）；Nanyang Technological University（南洋理工大学）；Chongqing University of Posts and Telecommunications（重庆邮电大学）
Vision-and-Language Training Helps Deploy Taxonomic Knowledge but Does Not Fundamentally Alter It
Boston University（波士顿大学）；University of Amsterdam（阿姆斯特丹大学）；Umeå University（乌梅大学）；Vrije Universiteit Amsterdam（阿姆斯特丹自由大学）；TTIC（TTIC研究中心）；The University of Texas at Austin（德克萨斯大学奥斯汀分校）
Blind to Position, Biased in Language: Probing Mid-Layer Representational Bias in Vision-Language Encoders for Zero-Shot Language-Grounded Spatial Understanding
Korea Advanced Institute of Science and Technology (KAIST)（韩国科学技术院）；Samsung Electronics（三星电子）
T-Rex: Task-Adaptive Spatial Representation Extraction for Robotic Manipulation with Vision-Language Models
School of Software Engineering, South China University of Technology（软件工程学院，华南理工大学）；School of Future Technology, South China University of Technology（未来技术学院，华南理工大学）；Shien-Ming Wu School of Intelligent Engineering, South China University of Technology（智能工程学院，华南理工大学）
Eyla: Toward an Identity-Anchored LLM Architecture with Integrated Biological Priors -- Vision, Implementation Attempt, and Lessons from AI-Assisted Development
