Statistics and neural network mid-level books

April 26, 2016 6:43 AM

I have a computer science background, but never got much formal training in statistics. I'm working more with data from psychological experiments and am interested in neural networks to model human behavior. I'm looking for two books (or maybe more): an intermediate book on statistics focused on applications in psychology and a basic book in modern neural networks.

I'm familiar with the basics of practical (t tests, ANOVA), but I'm a little lacking on the theory and feel unsure on questions like when is a certain test applicable and what assumptions are made for which tests. I've been using R, so I would prefer a book that uses that for examples rather than another statistics package.

For the neural networks book, I have some experience in machine learning but it seems like neural networks are much more complicated. I would like a book that balanced some theory with practical use. I'm interested in using neural networks for modeling, not to implement or develop my own neural network package.

posted by demiurge to Science & Nature (2 answers total)

Best answer: Goodfellow et al.'s Deep Learning is likely to become the standard academic graduate level reference. It's was recently completed and is currently freely available online. It's also extremely dense and comprehensive, and probably not the place to start.

The places to start are:

• Michael Nielsen's short online book, which has short, library-free python listings that will help you build up some intuition for what's going on during train and predict, before you start using the batteries-included libraries like keras, etc. I don't love the notation in this book (and I'm a physicist, so I've got a strong stomach), but it's a great place to start.
• If you like videos, do Andrew Ng's Machine Learning Coursera course. I cannot recommend it highly enough. The 2 weeks on neural networks are time extremely well spent (but they'll only make sense if you do the ~3 weeks before then on traditional machine learning).
• Chris Olah's essays
• And if you get into NLP, Yoav Goldberg's NLP primer and any papers from, e.g. the Stanford NLP reading club (a lot of the important results from the last 5 years or so are not settled enough to be in even very new textbooks).

One thing to note, given your stated interests: you won't see much mention of the brain or humans in these books. The "neural" in neural networks isn't completely unrelated to the human brain, but it's really just an analogy. If you're interested in building artificially intelligent systems then neural networks are a great place to work. But won't learn much about human behavior unless you're working at, e.g. the academic cutting edge of computer vision.
posted by caek at 8:04 AM on April 26, 2016 [3 favorites]

Check out David MacKay's free book? I remember it being pleasant to read and used as one of the textbooks in a course on neural networks, although this was a decade ago.
posted by L0 at 10:25 PM on April 30, 2016 [1 favorite]