Comprehensive machine learning text[book]?
November 5, 2013 7:33 AM   Subscribe

As a post-Masters student in a technical discipline -- computer-science-related -- I find myself wishing to fill in knowledge gaps from courses that I took in school. I'm looking for one, long, comprehensive text on Machine Learning (or several shorter texts) from which to self-teach the whole ML process from beginning to end. Topics of interest would be - supervised, unsupervised and semi-supervised learning - clustering, classification etc. Having already delved into Manning's 'Foundations of Statistical Natural Language Processing' in school I feel I've a decent knowledge of SVMs, perceptrons, clustering, (H)MMs, etc. but I'd like to learn more about general algorithm design, ML and how it has evolved over the years. Thanks! :)
posted by code_n_cakes to Computers & Internet (4 answers total) 25 users marked this as a favorite
 
My outrageously opinionated thoughts:

The best text for history is AIMA.

The best text for learning is Pattern Recognition and Machine Learning, or the Bishop book.

Ullman, Rajaraman and Leskovic have a data mining book that's really good.

PGM's are definitely best taught by Koller and Friedman.

Here's a RL book.

Smola has a fairly fast and quick introduction.

Andrew Ng's lecture notes might as well be a textbook.

Two deep learning tutorials of note are Ng's and Bengio's.

If you like math a lot, then the Elements is for you.
posted by curuinor at 8:37 AM on November 5, 2013 [5 favorites]


Excellent and comprehensive outline. I was going to talk about the Bishop (using it for a class right now), though you'll get a heavy dose of Bayes with just about every topic.

Still, for mathematical foundations of the techniques, it's pretty good.

Echoing the Ng lecture notes-- I find myself there a lot for another approach to a topic in my class, and it's never been unhelpful.
posted by supercres at 10:50 AM on November 5, 2013


Ng is in fact teaching the Machine Learning class that started Coursera again right now, and you "missed" officially only the first week of material (which are two hours of lectures that are still available and one homework you might be able to turn in late.)

I don't know how comprehensive it will be, and it's not very high level probably, but at least it's going to cover the landscape and probably have links to further resources.
posted by seyirci at 3:15 PM on November 5, 2013


Ng is a cool person and basically the most overworked person I have ever seen or probably will ever see in my life, interestingly enough. The Coursera class is not the best. The Stanford Engineering Everywhere course is better, and more rigorous. But I would use the Bishop book in conjunction with it, although Bishop's notation is a tad different than Ng's notation.
posted by curuinor at 3:46 PM on November 5, 2013


« Older Name that blog!   |   Papers, papers everywhere, but not the one I'm... Newer »
This thread is closed to new comments.