In this talk, I present a selection of ideas and algorithms that are presented in my recently published textbook of the same title [1]. The book introduces information measurement methodologies for machine learning that reduce the reliance on hyperparameters and model-type biases. This information-driven perspective enables data quality measurements, task complexity estimations, and reproducible experimental design. Bridging the disciplines of machine learning, information theory, and computer engineering, the methods presented enhance explainability and model resilience, contributing to the robustness and credibility of data science as an engineering discipline. The textbook evolved over many years out of a computer science graduate seminar in combination with the "Information and Uncertainty" discussion group at BIDS.
373