Abstract:
Explainable Artificial Intelligence (XAI) is a very active domain that has emerged from the increased complexity in nowadays data analysis pipelines. The objective is generally to discover the most likely causes for observed results or to make a data analysis process more interpretable. In that sense, the explanation process relates to iterative hypotheses expressed by a user to gain a better understanding of the data and how these data are used by models. Explaining can thus be formalized as a specific Exploratory Data Analysis (EDA) task that more generally provides tools and algorithms to help users in the tedious task of understanding their data and extract meaningful insights. In this talk, I will briefly first describe some contributions related to the explanation of recommendations, then presents some works related to automated EDA that draws inspiration from Operation Research to extract insights and finally illustrates how explanations can help exploring hypotheses on data.