Lisa_seminaires Mars 2011

lisa_seminaires@iro.umontreal.ca

2 participants
6 discussions

Fwd: (reminder) UdeM-McGill-MITACS machine learning seminar Wed March 16th @ 14h00, Z-345
by Guillaume Desjardins 15 Mar '11

15 Mar '11

A reminder for tomorrow's MITACS talk by Brendan Frey. Note that the seminar is again moving to another room, Z-345 (McNicoll Building). ---------- Forwarded message ---------- From: Guillaume Desjardins <guillaume.desjardins(a)gmail.com> Date: Thu, Mar 10, 2011 at 7:41 PM Subject: Another (!) UdeM-McGill-MITACS machine learning seminar Wed March 16th @ 14h00, Z-345 To: lisa_seminaires(a)iro.umontreal.ca Another UdeM-McGill-MITACS machine learning seminar will be held this Wednesday, March 16th. The talk given by Brendan Frey, will take place from 14h00-15h00 in the room Z-345 (McNicoll building) at the Université de Montréal. Hope to see you there ! Title: Learning models of transformed image parts Speaker: Brendan Frey Abstract: Natural signals such as images and audio can be described as a composition of transformed features. Transformations such as translation and scaling relate different parts of the same object, and relate different instances of the same object drawn from different training cases. In this talk, I'll review generative models that account for transformations in space, time and signal intensity. These transformations correspond to changes in shape, motion and color. Then, I'll describe my group's current efforts to build hierarchical models that relate transformations of object parts.

1 0

Fwd: (reminder) UdeM-McGill-MITACS machine learning seminar Mon March 14th @ 14h00, Z-200
by Guillaume Desjardins 14 Mar '11

14 Mar '11

A reminder for today's MITACS talk by Guido Montufar. Note that the seminar is NOT held in the usual location and will take place in the McNicoll Building, room Z-200. ---------- Forwarded message ---------- From: Guillaume Desjardins <guillaume.desjardins(a)gmail.com> Date: Thu, Mar 10, 2011 at 7:37 PM Subject: UdeM-McGill-MITACS machine learning seminar Mon March 14th @ 14h00, Z-200 To: lisa_seminaires(a)iro.umontreal.ca A UdeM-McGill-MITACS machine learning seminar will be held this Monday, March 14th. The talk given by Guido Montúfar, will take place from 14h00-15h00 in the room Z-200 (McNicoll building) at the Université de Montréal. Hope to see you there ! Title: On Exponential Families and Expressive Power of Related Generative Models Speaker: Guido Montúfar Abstract: In this talk I will present results on exponential families which give information on the representational power of diverse generative models. Generative models are able to generate high order correlations among visible variables even when demanding a series of conditional independence conditions (important for efficient learning). The question is: how many hidden variables must they contain? Deep Belief Networks (DBNs) and Restricted Boltzmann Machines (RBMs) generate weighted sums of conditional distributions, which means they are special kinds of mixture models. I will discuss a scheme for decomposing probability distributions as mixtures of exponential families and from this derive some results for DBNs and RBMs.

1 0

Another (!) UdeM-McGill-MITACS machine learning seminar Wed March 16th @ 14h00, Z-345
by Guillaume Desjardins 10 Mar '11

10 Mar '11

Another UdeM-McGill-MITACS machine learning seminar will be held this Wednesday, March 16th. The talk given by Brendan Frey, will take place from 14h00-15h00 in the room Z-345 (McNicoll building) at the Université de Montréal. Hope to see you there ! Title: Learning models of transformed image parts Speaker: Brendan Frey Abstract: Natural signals such as images and audio can be described as a composition of transformed features. Transformations such as translation and scaling relate different parts of the same object, and relate different instances of the same object drawn from different training cases. In this talk, I'll review generative models that account for transformations in space, time and signal intensity. These transformations correspond to changes in shape, motion and color. Then, I'll describe my group's current efforts to build hierarchical models that relate transformations of object parts.

1 0

UdeM-McGill-MITACS machine learning seminar Mon March 14th @ 14h00, Z-200
by Guillaume Desjardins 10 Mar '11

10 Mar '11

A UdeM-McGill-MITACS machine learning seminar will be held this Monday, March 14th. The talk given by Guido Montúfar, will take place from 14h00-15h00 in the room Z-200 (McNicoll building) at the Université de Montréal. Hope to see you there ! Title: On Exponential Families and Expressive Power of Related Generative Models Speaker: Guido Montúfar Abstract: In this talk I will present results on exponential families which give information on the representational power of diverse generative models. Generative models are able to generate high order correlations among visible variables even when demanding a series of conditional independence conditions (important for efficient learning). The question is: how many hidden variables must they contain? Deep Belief Networks (DBNs) and Restricted Boltzmann Machines (RBMs) generate weighted sums of conditional distributions, which means they are special kinds of mixture models. I will discuss a scheme for decomposing probability distributions as mixtures of exponential families and from this derive some results for DBNs and RBMs.

1 0

Tea Talk Tomorrow -- NOTE: CHANGE OF TIME
by Aaron Courville 09 Mar '11

09 Mar '11

Hi All, Tomorrow we will have another tea talk from the LISA lab members at the core of our UTLC team. Note the change of time! Hope to see you there. Location: LISA lab (AA3256) Time: 13h00 Date: Thursday, March 10th, 2011 Title: Unsupervised and Transfer Learning Challenge (UTLC) Authors: Yann, Xavier, Salah, Grégoire Abstract: The phase 1 of unsupervised learning challenge aimed to test unsupervised learning methods on five different datasets (sparse/dense, high/low dimension, images/manuscripts/... recognition). Competitors were evaluated with respect to the ALC (Area under Learning Curve) of a linear classifier (Hebbian) trained with very few training examples on a subset of the classes present in the training set. This allowed us to test deep/shallow learning algorithms in new settings: one shot learning (one sample used for training) and transductive learning. We will present the different methods we used to achieve good performance (ranking 1st overall on the validation set) and our positive/negative feedback about the competition. Cheers, Aaron -- Aaron C. Courville Département d’Informatique et de recherche opérationnelle Université de Montréal email:Aaron.Courville@gmail.com

1 0

UdeM-McGill-MITACS machine learning seminar Fri March 4th @ 14h00, AA3195
by Guillaume Desjardins 03 Mar '11

03 Mar '11

A UdeM-McGill-MITACS machine learning seminar will be held this Friday, March 4th. The talk given by Hugo Larochelle, will take place from 14h00-15h00 in the room AA3195 (Université de Montréal). Hope to see you there ! Title: Two new autoencoders for distribution estimation and guided representation, Hugo Larochelle (University of Toronto) Abstract: In this talk, I'll describe two new autoencoder-like models, developed for two different problems. The first is the estimation of distributions of high-dimensional data, for which the restricted Boltzmann machine (RBM) has been shown to be a powerful model. However, an RBM typically does not provide a tractable distribution estimator, since evaluating the probability it assigns to some given observation requires the computation of the partition function, which itself is usually intractable. The model I'll describe circumvents this difficulty by decomposing the joint distribution of observations into tractable conditional distributions and modeling each conditionals with a non-linear function similar to RBM conditionals. This model can also be interpreted as an autoencoder wired such that its output can be used to assign valid probabilities to observations. I'll present experiments showing that this new model outperforms other multivariate binary distribution estimators on several datasets and performs similarly to a large (but intractable) RBM. The second problem is that of guiding an autoencoder toward representations that are more useful for particular discriminative tasks. A complementary challenge is finding codes that are explicitly invariant to irrelevant transformations of the data. I'll describe how this can be achieved by combining an autoencoder with a Gaussian process latent variable model. This enables an autoencoder's unsupervised representation to both incorporate relevant label information and ignore irrelevant variations. I'll describe experiments on several different datasets which show how both labels and nuisance variables can provide cues for useful latent representations. This is joint work with Iain Murray, Jasper Snoek and Ryan Prescott Adams.

1 1

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

Lisa_seminaires Mars 2011