Stability of Tikhonov Regularization
Lorenzo Rosasco and Tomaso Poggio

Description

We briefly review the generalization bounds of last lecture before turning to our main goal -- using the stability approach to prove generalization bounds for Tikhonov regularization in RKHS. In order to apply the bounds, we need to prove that Tikhonov regularization is uniformly stable with beta=O(1/n), and also to bound the loss function. In the process, we will gain additional insight into the mathematics of optimization and RKHS.

Slides

Slides for this lecture: PDF

Suggested Reading

• O. Bousquet and A. Elisseeff. Stability and Generalization. Journal of Machine Learning Research, 2002.