We combine the replica approach from statistical physics with a variational approach to analyze learning curves analytically. We apply the method to Gaussian process regression. As a main result we derive approximative relations between empirical error measures, the generalization error and the posterior variance.
|Journal||Advances in Neural Information Processing Systems|
|Publication status||Published - Sep 2002|
Bibliographical noteCopyright of Massachusetts Institute of Technology Press (MIT Press). Available from Google Scholar.
- Gaussian process regression
- generalization error
- posterior variance