Directly Regularised Predictors

Avraham Ruderman

CS HDR MONITORING

DATE: 2012-10-29
TIME: 09:00:00 - 09:30:00
LOCATION: NICTA - 7 London Circuit
CONTACT: JavaScript must be enabled to display this email address.

ABSTRACT:
A directly regularised predictor is a linear predictor with kernel evaluations centred at data points as features. Examples of this are the LP machine and Kernel LASSO. These predictors have the advantage of not being formulated as an RKHS predictor, thus offering more flexibility when choosing the form of regularisation. I will present some new theoretical results on directly regularised estimators as well as some open practical and theoretical questions.