Bayesian Concept Bottleneck Models with LLM Priors

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Feng, Jean, Kothari, Avni, Zier, Luke, Singh, Chandan, Tan, Yan Shuo
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866917122802712576
author Feng, Jean
Kothari, Avni
Zier, Luke
Singh, Chandan
Tan, Yan Shuo
author_facet Feng, Jean
Kothari, Avni
Zier, Luke
Singh, Chandan
Tan, Yan Shuo
contents 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 exploring a sufficiently large set of concepts versus controlling the cost of obtaining concept extractions, resulting in a large interpretability-accuracy tradeoff. 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. Even though LLMs can be miscalibrated and hallucinate, we prove that BC-LLM can provide rigorous statistical inference and uncertainty quantification. Across image, text, and tabular datasets, BC-LLM outperforms interpretable baselines and even black-box models in certain settings, converges more rapidly towards relevant concepts, and is more robust to out-of-distribution samples.
format Preprint
id arxiv_https___arxiv_org_abs_2410_15555
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Bayesian Concept Bottleneck Models with LLM Priors
Feng, Jean
Kothari, Avni
Zier, Luke
Singh, Chandan
Tan, Yan Shuo
Machine Learning
Artificial Intelligence
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 exploring a sufficiently large set of concepts versus controlling the cost of obtaining concept extractions, resulting in a large interpretability-accuracy tradeoff. 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. Even though LLMs can be miscalibrated and hallucinate, we prove that BC-LLM can provide rigorous statistical inference and uncertainty quantification. Across image, text, and tabular datasets, BC-LLM outperforms interpretable baselines and even black-box models in certain settings, converges more rapidly towards relevant concepts, and is more robust to out-of-distribution samples.
title Bayesian Concept Bottleneck Models with LLM Priors
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2410.15555