Model Explainability Forum - #401
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
00:00
Evaluating Synthetic Neighborhoods and Perturbation Vulnerabilities
This chapter explores the challenges of assessing synthetic neighborhoods created during post hoc explanations in machine learning. It discusses issues around validating data perturbations, the limitations of finite samples, and the vulnerabilities these methods may pose to minority groups in datasets.
Transcript
Play full episode
