The Geometry of Representational Failures in Vision Language Models
For researchers studying VLM interpretability and reliability, this work offers a mechanistic explanation for common visual errors, though it is incremental as it applies known geometric analysis to VLMs.
The paper identifies that geometric overlap between concept vectors in VLMs correlates with specific error patterns like hallucination and misidentification, providing a quantitative framework to understand visual failures.
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.