Can explanations be trusted Pan Kessel Py Data Berlin August 2020

Abstract: Explanation methods in Machine Learning are on the rise. This is unsurprising as they promise to provide a tool to make blackbox algorithms transparent. This, in turn, can lead to increased trust and reliability. Furthermore, explanation methods are very simple to deploy as they are now integrated in standard deep learning libraries. In this talk, I will however demonstrate that explanations have to be considered with care. This is because they can be easily manipulated to closely reproduce an al