Talk by David Lopez-Paz (visitor from Ghahramani and Schölkopf's groups)
, University of Cambridge and MPI for Intelligent Systems
Date: Friday, December 13, 2013
Time: 1:00 PM to 2:00 PM Note: all times are in the Eastern Time Zone
Location: G4 Lounge (Stata)
Host: David Reshef, CSAIL
Contact: David Reshef, email@example.com
Relevant URL: http://people.tuebingen.mpg.de/dlopez/
Speaker URL: None
TALK: David Lopez-Paz (visitor from Ghahramani and Schölkopf groups)
Title: Using Randomness to Discover Patterns in the Large-Scale
Speaker: David Lopez-Paz, MPI for Intelligent Systems and University of Cambridge (lopezpaz.org)
Abstract: The Random Projection method is quickly gaining popularity in the Machine Learning Community. Of special importance is the seminal work of Rahimi and Recht (2008), who showed that random non-linear features may be used in conjuction with linear models to perform tasks such as non-linear regression or classification. These randomized methods show great performance and robustness, while achieving dramatic computational savings.
I will introduce the Randomized Dependence Coefficient (RDC), a measure of non-linear dependence between random variables of arbitrary dimension based on the Hirschfeld-Gebelein-Renyi Maximum Correlation Coefficient. RDC is defined in terms of correlation of random non-linear copula features; it is invariant with respect to marginal distribution transformations, has low computational cost and is easy to implement: just five lines of R code.
Some unpublished results on Randomized Component Analysis (PCA, CCA) will be also presented.
Joint work with Philipp Hennig and Bernhard Schoelkopf.
Created by David Reshef at Thursday, December 12, 2013 at 3:38 PM.