Auditing Black-box Models

Explore a hands-on tutorial from the FAT* 2018 conference that delves into auditing black-box models. Learn techniques to interpret complex machine learning models with multiple inputs and parameters. Discover how to identify potential indirect dependencies on protected attributes and assess the importance of various input factors. Follow along with the presenters as they guide you through a series of simple examples using a Jupyter notebook and their custom software library. Gain insights into the current capabilities and limitations of this auditing method, and understand how to apply it to your own datasets. This 57-minute session, led by experts from the University of Arizona, University of Utah, and Haverford College, equips you with practical skills to enhance model transparency and fairness in your machine learning projects.