Disentangling Hallucinations: Orthogonal Semantic Projection for Robust Interpretability
For users of XAI methods in safety-critical applications, this work addresses a fundamental flaw in attribution maps for Vision-Language Models.
The paper identifies and formalizes semantic hallucination in XAI for Vision-Language Models, showing it stems from Linear Semantic Leakage. It proposes OSP, a geometric intervention that reduces hallucination by orthogonalizing query vectors, achieving a 40% reduction in hallucination while preserving fidelity.
As Vision-Language Models are increasingly deployed in safety-critical applications, the trustworthiness of their explanations becomes crucial. Explainable AI (XAI) methods for Vision-Language Models often suffer from semantic hallucination, where attribution maps highlight prominent image regions even when prompted with incorrect text descriptions (e.g., highlighting a dog when prompted ``cat''). Although this problem is widespread, a formal mathematical analysis of XAI methods and CLIP embeddings is largely missing in the literature. We demonstrate that this phenomenon is not specific to a single architecture but is a fundamental consequence of Linear Semantic Leakage in high-dimensional embedding spaces. We propose a unified theoretical framework, Linear Semantic Attribution (LSA), which generalizes across discriminative methods. We introduce OSP, a geometric intervention that utilizes the residual property of OMP to disentangle unique semantic signals from shared concepts. We prove theoretically and demonstrate empirically that OSP minimizes hallucination by orthogonalizing the query vector against distractor concepts, rendering the attribution model blind to shared features while preserving fidelity for correct prompts. Our code is available at: https://github.com/emirhanbilgic/Orthogonal-Semantic-Projection