Please rec some good learning materials

May 14, 2012 8:14 AM

Statistics, machine learning, and image analysis/processing. Two months. Self-study. No other obligations. Recommendations?

I need to get a firm grasp on all three topics before graduate school and would really appreciate suggestions for learning materials! I have a computer science degree with a math minor (no, stats was not required), so the sky is the limit! "Out-of-the-box" suggestions are very welcome. Feel free to comment on one, two, or all. I won't be in a classroom setting, so anecdotes about your lightbulb learning experiences are good!

posted by puppetsock to education (10 answers total)

Students in my Coursera NLP class have had good things to say about the Stanford machine learning class there.
posted by yerfatma at 8:40 AM on May 14, 2012

People have said they've had good luck running through the Khan Academy's statistics videos. A glance seems to suggest they cover most of the topics a normal stats 101 course would cover.
posted by General Malaise at 9:06 AM on May 14, 2012 [1 favorite]

As a graduate student in cognitive science, I'm going to tell you that you're not going to get more than a cursory intro to any of those topics in two months.

Statistics is a huge area and what you need to know right off the bat will largely depend on the type of data you plan to work on. Maybe you want to read an intro book on bootstrap/Monte Carlo simulation? I actually think that is an excellent way to enter the world of stats because you can use it on any problem, it's very flexible as an analysis technique and it gets right down to the underlying logic of hypothesis testing. However, graduate level stats is something that takes a few semesters to really sink in. Just the very intro stuff, learning t-tests and simple linear regression is usually a semester. You can probably do it in two months, but get your hands on some software like R (which is free) or Matlab, which you need to pay for. Figure out what your department/lab uses and start playing with that. Get familiar with how to import data and do basic tests, and you'll be at least a half a semester ahead of the game when you start. Rice also has a nice Virtual Stats Lab.

Image analysis- this could mean any number of things. I'd need more clarification, but if you mean fMRI, I can definitely make recommendations about that. I'm not really qualified to comment on the machine learning stuff either.
posted by slow graffiti at 9:09 AM on May 14, 2012

A good intro to image processing/analysis is The Image Processing Handbook by John C. Russ. It's less mathematically rigorous than some books in the field, but it's a great "user's manual" if you are going to be writing image processing algorithms and need a broad toolbox of available techniques. Note that it doesn't really get into computer vision, it's mainly in the 2D realm. The techniques in this book are not cutting edge but they are very widely used.
posted by scose at 9:24 AM on May 14, 2012

Pattern Recognition and Machine Learning by Christopher Bishop.
posted by logicpunk at 9:46 AM on May 14, 2012

I found the parts of The Elements of Statistical Learning that I read to be pretty informative. Bonus: the authors of the book distribute the book for free in PDF form on their website.
posted by mean square error at 12:17 PM on May 14, 2012 [1 favorite]

By any chance, are you doing Hacker School?
posted by brainwane at 1:23 PM on May 14, 2012

All of Statistics is terse but a good intro to statistics, particularly the type you need to do machine learning.
posted by en forme de poire at 3:30 PM on May 14, 2012

I need to get a firm grasp on all three topics before graduate school

In addition to this being impossible, there is no program in the world which actually expects that. Email teachers for your upcoming G1 classes and ask the same question; they can tell you specifically what you need.
posted by a robot made out of meat at 5:25 PM on May 14, 2012

BTW, I have actually used "All of Statistics" as a self-study book. One summer my friends and I sort of made up a fake class, using it as the "textbook." Basically, we divided up the chapters and alternated giving lectures to the others and assigning homework problems, which we compared notes on at the start of the next session. This worked really well, especially for the times I had to "teach." If I were doing it again I would probably progress more slowly through the first several chapters, rather than trying to cover 2-3 per week.
posted by en forme de poire at 9:44 PM on May 14, 2012