Gaussian Process Regression


What's that?

  • GP regression uses GPs for function regression
  • a GP is a stochastic process, that is fully specified by a mean and a covariance function
  • the covariance function is specified by some kernel funktion k(t,t')
  • it specifies how the function values f(t) and f(t') can change, for given arguments t and t'
  • with this, a GP defines us a family of functions, not just one
  • we can draw randomly samples from such a GP to see examples of functions
  • now for GP regression you have some example measurements of your function
  • given by {t1,…,tn} and {y1=f(t1),…,yn=f(tn)}
  • you can now compute the mean- and covariance-function of a new GP that is restricted to that measurements
  • that is what GP regression is all about
  • the new GP will have small variance of function values nearby to the measurements and larger far away from that measurements

Helpful slides

Videos by mathematicalmonk

GP regression as a generalization of Bayesian linear regression

Derivation of the distribution for the posterior probability

Actual explanation of GP regression

Videos to be watched next

Video #1

public/gaussian_process_regression.txt · Last modified: 2014/01/18 10:24 (external edit) · []
Recent changes RSS feed Powered by PHP Valid XHTML 1.0 Valid CSS Driven by DokuWiki