Explainable Artificial Intelligence: An Introduction to Interpretable Machine Learning

In this comprehensive book on explainable AI (XAI) by John Liu and Uday Kamath, we find a valuable and thorough handbook for the AI/ML community, including early learners, experienced practitioners, and researchers. The book
covers the various dimensions of the AI technical risk taxonomy, including methods and applications. The result of the authors’ extensive work is an in-depth coverage of many XAI techniques, with real-world practical examples including code, libraries, algorithms, foundational mathematics, and thorough explanations (the essence of XAI).
The XAI techniques presented in-depth here range from traditional whitebox (explainable) models (e.g., regression, rule-based, graphs, network models) to advanced black-box models (e.g., neural networks). In the first case, XAI is
addressed through statistical and visualization techniques, feature exploration and engineering, and exploratory data analysis. In the latter case, XAI is addressed through a remarkably powerful and rich set of methods, including feature sensitivity testing through dependence, perturbation, difference, and gradient analyses. Extra attention is given to three special cases for XAI: natural language processing (NLP), computer vision (CV), and time series.
The book aims (and succeeds) to bring the reader up to date with the most modern advances in the XAI field, while also giving significant coverage to the history and complete details of traditional methods. In all the examples, use cases, techniques, and applications, the consistent theme of the book is to provide a comprehensive overview of XAI as perhaps the most critical requirement in AI/ML both for now and in the future, from both technical and compliance perspectives. In that regard, the forward-looking discussions at the end of the book give us a glimpse
of emerging XAI research areas that will advance our AI risk-compliance posture, including human-machine collaboration (as in assisted and augmented intelligence applications), causal
…