Gaussian processes are an interesting set of techniques in the statistics and machine learning repertoire. They are flexible regressors/classifiers and they offer a Bayesian measure of uncertainty, unlike other flexible predictors. Somewhat more theory is required to get started with Gaussian processes than with other supervised learners, but the extra inferential utility makes up for this burden. Those who desire strong theoretical foundations for their methods may actually find that requirement appealing.
Gaussian Processes for Machine Learning
Carl E. Rasmussen and Christopher K. I. Williams
- In-text exercises
- Errata, code, and full .pdf
This is the canonical book on Gaussian processes in the machine learning community. It's somewhat terse, but it does have a number of positive things going for it: there aren't many other options, it comes with code (Matlab unfortunately), and the authors provide a free electronic copy of the book. The exercises are rather theoretical for a machine learning book, but you can gain a lot of insight by modifying the provided code or reproducing it in your language of choice. This website lists a lot of Gaussian process related software packages and other resources, which may be useful to you if you decide you're interested in the topic.