---------- Forwarded message ----------
From:
Guillaume Desjardins <guillaume.desjardins@gmail.com>
Date: Fri, Jul 15, 2011 at 5:31 PM
Subject: UdeM-McGill-MITACS machine learning seminar Wed July 20th @ 14h00, AA3195
To:
lisa_seminaires@iro.umontreal.ca
A UdeM-McGill-MITACS machine learning seminar will be held this Wesdnesday, July 20th. The talk given by Hugo Larochelle, will take place from 14h00-15h00 in the room AA3195 (pavillon Andre-Aisenstadt, click
here for directions) at the Université de Montréal. Hope to see you there !
Title: Restricted Boltzmann Machines with Structure: Two Extensions
Abstract:
In this talk, I'll describe two extensions of the restricted Boltzmann machine to classification problems that exhibit more structure than the standard setting with "input vector / target class index" pairs.
In the first extension, we'll consider the problem of classification when inputs correspond to sets of vectors. This setting occurs in many problems such as the classification of pieces of mail containing several pages, of web sites with several sections or of images that have been pre-segmented into smaller regions. I'll describe generalizations of the restricted Boltzmann machine (RBM) that are appropriate in this context and explore how to incorporate different assumptions about the relationship between the input sets and the target class within the RBM.
In the second extension, we'll look at the problem of structured output (target) prediction with an RBM. I'll discuss how contrastive divergence (CD) is not appropriate in this setting and describe two new learning algorithms that improve on CD.
The first extension is joint work with Jérôme Louradour, and the second extension is joint work with Volodymyr Mnih and Geoffrey Hinton.
--
Guillaume Desjardins