Dr. Sebastian Lapuschkin | XAI as a Tool Beyond Model Understanding: From Heatmaps to Concepts and XAI Automation

The field of eXplainable Artificial Intelligence (XAI) aims to bring transparency to powerful but opaque deep learning models. The vast majority of current approaches to XAI still provide very limited insights into the models' internal "reasoning" and largely leave the burden of (mis)interpreting their outputs to the user. This talk will focus on the novel approaches called Concept Relevance Propagation (CRP) and Prototypical Concept-based Explanation (PCX) to provide an outlook on how recent developments toward concept-based XAI may lead to more human-interpretable explanations, enabling novel ways to analyze models and data, with a particular focus on the human labor vs automation involved in the adaptation of XAI.