machine learning @ toronto
[ home ] [ people ] [ projects ] [ courses ] [ meetings ]


A critical view of some Gaussian Process implementations


Monday, January 28th 2008 — Iain Murray


Abstract:
 

Gaussian Processes offer a flexible framework for Bayesian non-parametric regression. This makes them appealing, until one is stuck trying to invert Gigabytes of covariance matrix. The talk will offer no new algorithms. Instead I will describe existing methods that work, and those that — despite papers written to the contrary — demonstrably don’t.