If Concept Bottlenecks are the Question, are Foundation Models the Answer?

2025-04-28

AI Summary

This study investigates leveraging vision-language models (VLMs) such as CLIP as weak supervision to replace costly expert annotations in constructing concept bottleneck models (CBMs), and systematically evaluates its impact on concept quality. Method: We propose a VLM-based weakly supervised CBM architecture, and introduce the Concept Quality Assessment (CQA) metric alongside a cross-task consistency analysis framework. Contribution/Results: We empirically demonstrate—for the first time—that concept accuracy under VLM supervision exhibits no strong correlation with downstream task performance, challenging the prevailing CBM evaluation paradigm by revealing that “high accuracy ≠ high explanatory validity.” Experiments show VLM supervision can improve concept accuracy by up to 12%, yet reduces explanation fidelity by up to 37% and induces significant distributional deviation from expert-annotated concepts. We open-source the CQA toolkit, establishing a reproducible benchmark for weakly supervised CBM research.

Technology Category

• Computer Vision: Large Vision Models
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Large pretrained models with web data
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

Concept Bottleneck Models (CBMs) are neural networks designed to conjoin high performance with ante-hoc interpretability. CBMs work by first mapping inputs (e.g., images) to high-level concepts (e.g., visible objects and their properties) and then use these to solve a downstream task (e.g., tagging or scoring an image) in an interpretable manner. Their performance and interpretability, however, hinge on the quality of the concepts they learn. The go-to strategy for ensuring good quality concepts is to leverage expert annotations, which are expensive to collect and seldom available in applications. Researchers have recently addressed this issue by introducing"VLM-CBM"architectures that replace manual annotations with weak supervision from foundation models. It is however unclear what is the impact of doing so on the quality of the learned concepts. To answer this question, we put state-of-the-art VLM-CBMs to the test, analyzing their learned concepts empirically using a selection of significant metrics. Our results show that, depending on the task, VLM supervision can sensibly differ from expert annotations, and that concept accuracy and quality are not strongly correlated. Our code is available at https://github.com/debryu/CQA.

Problem

• Evaluating impact of foundation models on concept quality in CBMs
• Comparing VLM supervision versus expert annotations in CBMs
• Assessing correlation between concept accuracy and quality in CBMs

Innovation

• Using foundation models for weak supervision
• Replacing expert annotations with VLM-CBM
• Analyzing concept quality with empirical metrics