Informative Perturbation Selection for Uncertainty-Aware Post-hoc Explanations
The episode discusses the paper 'Informative Perturbation Selection for Uncertainty-Aware Post-hoc Explanations.' The discussion introduces EAGLE, an active learning method that efficiently selects data perturbations to explain black-box AI models. This approach improves explanation stability and provides confidence intervals, allowing users to understand not just what the model predicts, but how sure it is about that prediction.