1
$\begingroup$

I am using the MATLAB code for Rasmussen & Williams' book Gaussian Processes for Machine Learning.

How can one incorporate prior knowledge in Gaussian process regression? Say, that the variance in one dimension of a two dimensional vector is greater. Is it only by considering the parameters of a normal distribution, or can it be more detailed?

$\endgroup$
  • $\begingroup$ The prior is Gaussian. Do you have the actual book? If not, you can read the co-author's summary. Hanna Wallach also has an introductory presentation $\endgroup$ – Emre May 22 '12 at 3:53
  • $\begingroup$ Then it is not really a prior knowledge ! If I already know that the influence of variable A is more than variable B, then how should I consider it ?! $\endgroup$ – user4581 May 23 '12 at 22:47
  • $\begingroup$ so basically it is not possible ?! $\endgroup$ – user4581 Jun 3 '12 at 12:59
1
$\begingroup$

Have a look at Gelman et.al. Bayesian Data Analysis (Third Edition), chapter 21 which has a Bayesian treatment of Gaussian processes. It also has good applied examples.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.