How relevant is this book today?
In 2008 or so, Steve Abney (Michigan, linguistics) suggested this book to me while we were discussing a meaning-is-use theory of semantic content. I went straight to the library to get it but their copy was out. Never did get it. Happy to have it now.
A classic, if slightly dated in some chapters, books on ml.
For something recent, I really enjoy Amazon’s interactive explain articles on machine learning:
https://mlu-explain.github.io/
It’s nice to see the computational medium for education change with the times.
It looks pretty good. The introduction, ANN and Bayesian chapters look just as relevant today as they would have been when this book was published, 25 years ago. I like that hypotheses testing is covered.
Chapter 9, Genetic Algorithms and Programming, was hot stuff back then, and research continues today, I went to an interesting seminar a few months ago given by a genetic programming researcher, but from the perspective of solving a ML problem, less useful to the beginner. If you have a regression of classification problem, few people will be trying genetic algorithms to solve it.
Chapter 10 - 12 is rule based systems, for example Prolog, which again was thought of being the future back then, and as much as I love using Prolog, it is not part of modern ML curriculum.
Chapter 13 is short chapter on RL, which is popular today (DeepMind).
All that said, if a beginner wants to pick one book, I would suggest ISLR2, also free and has a Edx course
https://www.statlearning.com/
https://www.edx.org/course/statistical-learning
RL ca. 1961: https://en.wikipedia.org/wiki/Matchbox_Educable_Noughts_and_...
(which I learned about from a Fred Saberhagen short)
Is that the same one that was on the stanford edx clone site?
edx seems to be putting more behind a paywall and jacking up the price in a way that's pretty disappointing. That one is what $150 for the computer to tell you if the numbers you typed in some boxes is what was expected, for example. I mean, looking up the answer in the back of the book is worth that? The certificate is worth absolutely nothing to anyone save for how it makes them feel.
I really enjoy this book precisely because it focuses on many basics that are glossed over in more contemporary books.
For example, while every knows what a random forest is, I wonder how many of those people actually know how to design and implement a decision tree, the building blocks of random forests.
It also spends a lot more time with the theory behind machine learning, which many more contemporary books gloss over. Computational learning theory, and specifically topics like PAC learning rarely get the attention they should.
I would also like to add that no academic book in ML truly becomes "irrelevant" or "out of date". In 2007 someone asking same question might have criticized the book for spending too much time on Neural Networks since they aren't really relevant any more (if you're unaware, NNs were largely viewed as a limited novelty in light of more recent developments in SVMs and RandomForests in the early 2000s).
I don't know about this book, but there are August 2022 PDFs of "Probabilistic Machine Learning : An Introduction" and "Probabilistic Machine Learning : Advanced Topics" by Kevin Murphy available here: https://github.com/probml/pml-book
Is that recent enough?
It's very relevant.
The key innovation of machine learning was formalizing the process, not the specific algorithm. Mitchell's book does a superb job of illuminating that process across a diverse set of approaches.
It felt 'dated' 10 years ago.
Not that you can't use it. Intro ML books (like courses) are in any case generally too verbose and mathematically unsophisticated to cater to their audience.