Project P2 has two main tasks. These are (1) to enrich DL-based statistical tests with visual explanations and (2) to transfer the formalism of statistical tests into the domain of explainable artificial intelligence (XAI). The results of the first task will allow us to answer questions about a significant test, e.g. which features (e.g. pixels, words, genes) in a set of objects (e.g. images, documents, gene expressions) are dependent on a second variable (in the context of an independence test) or which features cause two sets of observations to differ in their distribution (in the context of a two-sample test). The result of the second task will provide novel statistical tests for the population analysis of a set of individual explanations computed with layer-wise relevance propagation (LRP).

The aim of this project is to provide visual explanations for significant tests, including the techniques developed in P1. The statistical tests for explanations developed in this project will be used in P3, P4, and P5.