Right for the Wrong Scientific Reasons: Revising Deep Networks by Interacting with their Explanations.
Patrick Schramowski, Wolfgang Stammer, Stefano Teso, Anna Brugger, Xiaoting Shao, Hans-Georg Luigs, Anne‐Katrin Mahlein, Kristian Kersting · arXiv (Cornell University) · 2020
Deep neural networks have shown excellent performances in many real-world applications. Unfortunately, they may show Hans-like behavior---making use of confounding factors within datasets---to achieve high performance. In this work we introduce the novel setting of explanatory interactive learning (XIL) and illustrate its benefits on a plant phenotyping research task. XIL adds the scientist into the training loop such that she interactively revises the original model via providing feedback on its explanations. Our experimental results demonstrate that XIL can help avoiding Clever Hans moments in machine and encourages (or discourages, if appropriate) trust into the underlying model.