FASTER-CE: Fast, Sparse, Transparent, and Robust Counterfactual Explanations

Counterfactual explanations were first introduced as a human-centric way to understand model behavior. While validity remains core to the counterfactual explanation definition, researchers have also identified other desirable properties that make counterfactual explanations more usable on the deployment and the end-user sides: speed of explanation generation, robustness/sensitivity, and succinctness of explanations. Motivated by the need to make counterfactual explanations practically viable for large-scale datasets, we introduce a novel set of algorithms called FASTER-CE, which generate sparse and robust counterfactual explanations efficiently at test time by finding promising search directions for counterfactuals in a latent space that is specified via an autoencoder. These directions are determined based on gradients with respect to each of the original input features as well as of the target, as estimated in the latent space. The ability to quickly examine combinations of the most promising gradient directions as well as to incorporate additional user-defined constraints enables FASTER-CE to generate multiple counterfactual explanations that are sparse, realistic, and robust to input manipulations. Through experiments on multiple datasets of varied complexities, we show that FASTER-CE is not only much faster at test time, but also capable of considering a larger set of desirable and often conflicting properties for counterfactual explanations.