Francesco Leofante will give a talk about his work on Robust Counterfactual Explanations for Deep Neural Networks. Details below.
SPEAKER
Francesco Leofante, Research Fellow, Imperial College.
TITLE
Robust Counterfactual Explanations for Deep Neural Networks
ABSTRACT
Deep Neural Networks (DNNs) are increasingly used for automated decision making. However, they often produce outputs that are not interpretable by humans, which limits their applicability in high-stakes scenarios. To remedy this, several approaches have emerged to generate explanations for DNNs, i.e., arguments supporting or contrasting their decisions. In this talk we will focus on counterfactual explanations (CFXs), an increasingly popular explanation strategy for DNNs. While several algorithms exist to generate such explanations, they are often lacking robustness, i.e., their validity may be compromised when small perturbations are applied to the DNN or to the CFX itself. In this talk we will introduce the problem of robustness, discuss its implications and present some recent solutions we developed to compute CFXs with formal robustness guarantees.