Statistics Seminars: An Identity for Kernel Ridge Regression
5 March 2012 14:00 in CM221
Ridge regression is a popular technique in machine learning and statistics with numerous applications. In the talk I will discuss ridge regression in the contexts of functional analysis (reproducing kernel Hilbert spaces) and the theory of random fields (Gaussian covariances) and derive an identity linking the quadratic losses of kernel ridge regression in batch and on-line frameworks. Some corollaries describing the behaviour of the cumulative loss of on-line ridge regression will be obtained. An alternative proof of the identity motivated by the aggregating algorithm will be presented.
The results of the talk are covered in the report "An Identity for Kernel Ridge Regression" by F.Zhdanov and Y.Kalnishkan (arXiv:1112.1390v1 [cs.LG]).
Contact firstname.lastname@example.org for more information