Less Wrong is a community blog devoted to refining the art of human rationality. Please visit our About page for more information.

alexflint comments on The Best Textbooks on Every Subject - Less Wrong

167 Post author: lukeprog 16 January 2011 08:30AM

You are viewing a comment permalink. View the original post to see all comments and the full post content.

Comments (327)

Sort By: Controversial

You are viewing a single comment's thread.

Comment author: alexflint 17 January 2011 11:12:14PM 4 points [-]

Machine learning: Pattern Recognition and Machine Learning by Chris Bishop

Good Bayesian basis, clear exposition (though sometimes quite terse), very good coverage of the most modern techniques. Also thorough and precise, while covering a huge amount of material. Compared to AI: A modern approach it is much more clearly based in Bayesian statistics, and compared to Probabilistic robotics it's much more modern.

Comment author: PhilGoetz 23 January 2011 08:18:32PM 8 points [-]

Bishop, vs Russell & Norvig, are not in the same category. There's only two chapters in R&N that overlap with Bishop.

Within the category of planning, symbolic AI, and agent-based AI, I recommend Russell & Norvig, "Artificial Ingelligence: A Modern Approach", or Luger & Stubblefield, "Artificial Intelligence". They are aware of non-symbolic approaches and some of the tradeoffs involved. I do not recommend Charniak & McDermott, "An intro to artificial intelligence", or Nilsson, "Principles of artificial intelligence", or Winston, "Artificial Intelligence", as they go into too much detail about symbolic techniques that you'll probably never use, like alpha-beta pruning, and say nothing about non-symbolic techniques. A more complete treatement of symbolic AI is Barr & Feigenbaum, "The Handbook of Artificial Intelligence", but that's a reference work, and I'm recommending textbooks. I do recommend a symbolic AI reference work, Shapiro, "Encyclopedia of Artificial Intelligence", because the essays are reasonably short and easy to read.

Within machine learning, data mining, and pattern recognition, I haven't read Bishop's work. Mannila & Smyth, "Principles of Data Mining", are often used; but maybe just because they're from MIT. Larose, "Data mining methods and models", is okay, as is its companion volumne whose name I forget. My favorite is Data Mining: Practical Machine Learning Tools and Techniques (Second Edition), by Ian H. Witten and Eibe Frank. It is brief, to the point, and gives coding examples using Weka.

The best advice I can give related to statistical modeling is to look up your technique in the SAGE series, and buy the SAGE books on it. They cost about $16 apiece, less used on amazon, and are short yet detailed. Now, I don't mean the books SAGE tries to sell you on their website. I mean the series of about 200 small light-green-cover paperbacks that they for some reason don't tell you about on their website.

But if you're reading this level of detail, it means you're going to be a specialist trying to implement or improve algorithms, and you're going to need to read entire books on each subject.