Robust Algorithmic Recourse with Predictions

This lecture from the Simons Institute features Shahin Jabbari of Drexel University discussing the concept of robust algorithmic recourse enhanced with predictions. Learn how machine learning models that provide individuals with suggestions for improvement can become invalid when models are updated over time. Explore a novel learning-augmented framework that aims to reduce the cost of recourse when predictions about future model changes are accurate while limiting costs when predictions fail. Understand the robustness-consistency trade-off in algorithmic recourse and discover how prediction accuracy affects overall performance. Part of the "Theoretical Aspects of Trustworthy AI" series, this talk introduces innovative approaches to ensure algorithmic suggestions remain valid despite model updates.