Bayesian Concept Bottleneck Models with LLM Priors

2024-10-21

arXiv.org

AI Summary

Traditional concept bottleneck models (CBMs) are constrained by predefined concept sets, leading to a fundamental trade-off between interpretability and accuracy. To address this, we propose the Bayesian Concept Bottleneck Model (BCBM), the first framework to embed large language models (LLMs) within a Bayesian inference paradigm—leveraging them as dynamic concept priors and generative concept extractors. BCBM enables automatic, provably convergent search over an infinite concept space while quantifying epistemic uncertainty. Our method integrates Bayesian inference, LLM-driven concept generation and evaluation, multimodal modeling, and sparse concept selection optimization. Experiments demonstrate that BCBM consistently outperforms both black-box models and conventional CBMs across diverse tasks. It achieves faster concept convergence, enhanced robustness to distributional shift, and provides statistically rigorous interpretability guarantees grounded in Bayesian posterior calibration.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Natural Language Processing: (Large) Language Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Concept Bottleneck Models (CBMs) have been proposed as a compromise between white-box and black-box models, aiming to achieve interpretability without sacrificing accuracy. The standard training procedure for CBMs is to predefine a candidate set of human-interpretable concepts, extract their values from the training data, and identify a sparse subset as inputs to a transparent prediction model. However, such approaches are often hampered by the tradeoff between enumerating a sufficiently large set of concepts to include those that are truly relevant versus controlling the cost of obtaining concept extractions. This work investigates a novel approach that sidesteps these challenges: BC-LLM iteratively searches over a potentially infinite set of concepts within a Bayesian framework, in which Large Language Models (LLMs) serve as both a concept extraction mechanism and prior. BC-LLM is broadly applicable and multi-modal. Despite imperfections in LLMs, we prove that BC-LLM can provide rigorous statistical inference and uncertainty quantification. In experiments, it outperforms comparator methods including black-box models, converges more rapidly towards relevant concepts and away from spuriously correlated ones, and is more robust to out-of-distribution samples.

Problem

• Overcoming interpretability-accuracy tradeoff in Concept Bottleneck Models
• Exploring infinite concept sets using Bayesian framework with LLMs
• Providing statistical inference despite LLM miscalibration and hallucination

Innovation

• Bayesian framework with LLM priors
• Iterative search over infinite concepts
• LLMs for concept extraction and prior