Reviews of AI and ML books by Marco Huber

S. Russell, P. Norvig: ‘Artificial Intelligence: A Modern Approach’, 2010, 3rd edition

This standard reference work on Artificial Intelligence was published in 1995 and has been continuously updated ever since. No other book combines depth and breadth with comprehensibility as well as this book. Only moderate mathematical knowledge is required and almost all subdisciplines of AI are covered, not just the area of machine learning as is the case with most other books. If you can only buy one book on AI, this is the one to get. The fourth edition has been announced for summer 2020.

Target groups: students of a master’s degree course in a STEM subject, engineers.

G. Strang: ‘Linear Algebra and Learning from Data’, 2019

Despite his advanced age of 85 years, Gilbert Strang has once again written a book on linear algebra. What does linear algebra have to do with AI? A lot, especially when it comes to machine learning. Alongside optimization theory, linear algebra is a key mathematical discipline that must be mastered if you want or need to delve deeper into AI. This is especially true for artificial neural networks, which Gilbert Strang focuses on in the last part of his book.

Target groups: students and users who are well versed in mathematics.

D. Spiegelhalter: ‘The Art of Statistics: Learning from Data’, 2019

Sir David Spiegelhalter taught Statistics at the University of Cambridge from the 1980s until his retirement. He wrote this book with the aim of conveying the subject of statistics in an interesting and understandable way, rather than in the usual dry style. And he has succeeded extraordinarily well. The various methods of statistics, which also include machine learning methods, are introduced at the beginning of each chapter with a practical example and developed to explain the respective method. Only very few equations are used, which is extremely unusual for a book on statistics.

Target groups: anyone who wanted to study statistics at some point but never dared to.

S. Russell: ‘Human Compatible: Artificial Intelligence and the Problem of Control’, 2019

Although many scientists dream of developing a form of AI that is at least equal or even superior to human intelligence, it is something that is feared by many people. Stuart Russell, co-author of the standard reference work ‘Artificial Intelligence: A Modern Approach’, addresses this dichotomy. It shows the far-reaching capabilities of today’s AI systems, but also what is still lacking in order to achieve superhuman AI. He also proposes an approach for developing future AI systems designed to serve mankind, thus avoiding the otherwise threatening conflict between man and machine. In contrast to the other books discussed here, this one deals with a look into the future and not with the fundamentals of AI.

Target groups: anyone interested in a critical but balanced approach to AI.

M. Mitchell: ‘Artificial Intelligence: A Guide for Thinking Humans’, 2019

Similar to ‘Human Compatible’ and ‘The Art of Statistics’, this book aims at being easy to understand. Melanie Mitchell chooses to combine state-of-the-art AI methods with personal stories and experiences. In a very comprehensible manner, she explains how today’s main methods such as deep neural networks, reinforcement learning or natural language processing work. She shows where these methods are successfully implemented, but also where their limits currently lie. Not a single equation is used. Instead, the complex content is successfully imparted thanks to her distinct talent for “storytelling”.

Target groups: AI rookies.

P. Domingos: ‘The Master Algorithm’, 2015

The AI sub-discipline of Machine Learning is further broken down into a broad range of trends. Pedro Domingos addresses the five main trends and briefly explains their main perspectives and methods. However, the focus of the book is primarily on the search for the “master algorithm”, i.e. an algorithm that combines all trends, in the same way that physics strives to find a universal equation that unites fundamental interactions. Domingos succeeds in finding a common superstructure for two machine learning trends. The rest remains “future work”.

Target groups: AI rookies.

S. Theodoridis: ‘Machine Learning: A Bayesian and Optimization Perspective’, 2015

This is intended for people who are looking for a book with mathematical depth, that derives a variety of machine learning processes in great detail and proves their features mathematically. However, it is not a book that you put on your bedside table in the evening or read through in one go. Rather, it is a rich reference work that lives from the author's didactic skills. Sergios Theodoridis succeeds like no other in demonstrating the connections between the various machine learning processes. With the recently published second edition, the section on Deep Learning is now also up to date.

Target groups: people with advanced AI knowledge, students and users with a good grasp of mathematics.

I. Goodfellow, Y. Bengio, A. Courville: 'Deep Learning' 2017

Among the numerous methods of machine learning, so-called “deep learning”, i.e. deep artificial neural networks, is the focus of most of today’s research and application work. Deep Learning is almost exclusively responsible for all the breakthroughs in the field of AI in recent years. The authors of this book are among the leading researchers in this area, and some of them have been so for decades. This is an in-depth introduction to the subject and requires some basic mathematical skills on the part of the reader. In return, readers are taught everything they need to know about Deep Learning.

Target groups: people with advanced AI knowledge, students and users well versed in mathematics.