Tell me why: Visual foundation models as self-explainable classifiers

2025-02-26

AI Summary

Current visual foundation models (VFMs) suffer from poor interpretability, while existing self-explaining methods either require backbone modification or rely on post-hoc explanation modules. To address this, we propose ProtoFM: a lightweight, end-to-end self-explaining framework that trains only a prototype-based classification head (≈1M parameters) atop frozen VFMs (e.g., ViT, CLIP), using a concept-weighted reconstruction loss and a faithfulness-aware objective. ProtoFM is the first to deeply integrate prototype learning with VFMs—without fine-tuning the backbone or adding post-hoc interpreters—thereby significantly improving explanation faithfulness. Experiments demonstrate that ProtoFM achieves classification accuracy on par with full fine-tuning across multiple benchmarks, while consistently outperforming state-of-the-art self-explaining models on standard interpretability metrics—including Infidelity and Deletion.

Technology Category

• Computer Vision: Interpretability, Explainability, and Transparency
• Machine Learning: Transparent, Interpretable, Explainable ML
• Humans and AI: Explainable AI (XAI) for Human Understanding

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Visual foundation models (VFMs) have become increasingly popular due to their state-of-the-art performance. However, interpretability remains crucial for critical applications. In this sense, self-explainable models (SEM) aim to provide interpretable classifiers that decompose predictions into a weighted sum of interpretable concepts. Despite their promise, recent studies have shown that these explanations often lack faithfulness. In this work, we combine VFMs with a novel prototypical architecture and specialized training objectives. By training only a lightweight head (approximately 1M parameters) on top of frozen VFMs, our approach (ProtoFM) offers an efficient and interpretable solution. Evaluations demonstrate that our approach achieves competitive classification performance while outperforming existing models across a range of interpretability metrics derived from the literature. Code is available at https://github.com/hturbe/proto-fm.

Problem

• Improve interpretability of visual foundation models
• Combine VFMs with prototypical architecture
• Enhance faithfulness of self-explainable classifiers

Innovation

• Combines visual foundation models
• Uses prototypical architecture
• Trains lightweight interpretable head