Extracting PAC Decision Trees from Black Box Binary Classifiers: The Gender Bias Case Study on BERT-based Language Models
Abstract
Decision trees are a popular machine learning method, known for their inherent explainability. In Explainable AI, decision trees can be used as surrogate models for complex black box AI models or as approximations of parts of such models. A key challenge of this approach is determining how accurately the extracted decision tree represents the original model and to what extent it can be trusted as an approximation of their behavior. In this work, we investigate the use of the Probably Approximately Correct (PAC) framework to provide a theoretical guarantee of fidelity for decision trees extracted from AI models. Based on theoretical results from the PAC framework, we adapt a decision tree algorithm to ensure a PAC guarantee under certain conditions. We focus on binary classification and conduct experiments where we extract decision trees from BERT-based language models with PAC guarantees. Our results indicate occupational gender bias in these models.
Cite
@article{arxiv.2412.10513,
title = {Extracting PAC Decision Trees from Black Box Binary Classifiers: The Gender Bias Case Study on BERT-based Language Models},
author = {Ana Ozaki and Roberto Confalonieri and Ricardo Guimarães and Anders Imenes},
journal= {arXiv preprint arXiv:2412.10513},
year = {2025}
}
Comments
This is a revision of the version published at AAAI 2025. We fixed an issue in Theorem 8 and run again all the experiments. We also fixed small grammar mistakes found while producing this revised version