Distributionally Robust Recourse Action

Duy Nguyen, Ngoc Bui, Viet Anh Nguyen

Keywords: Algorithmic recourse, Robust optimization

Abstract: Recourse actions, also known as counterfactual explanations, aim to explain a particular algorithmic decision by showing one or multiple ways in which the instance could be modified to receive an alternate outcome. Existing recourse recommendations often assume that the machine learning models do not change over time. However, this assumption does not always hold in practice because of data distribution shifts, and in this case, the recourse actions may become invalid. To redress this shortcoming, we propose the Distributionally Robust Recourse Action framework, which generates a recourse action that has high probability of being valid under a mixture of model shifts. We show that the robust recourse can be found efficiently using a projected gradient descent algorithm and we discuss several extensions of our framework. Numerical experiments with both synthetic and real-world datasets demonstrate the benefits of our proposed framework.

One-sentence Summary: We propose the framework of Distributionally Robust Recourse Action (DiRRAc) for designing a recourse action that is robust to mixture shifts of the model parameters.

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2302.11211/code)