September 2007 - Lisa_seminaires - mailman.iro.umontreal.ca

REMINDER: UdeM-McGill-MITACS machine learning seminar, PAA 1409 (UdeM), Fri Sept 28, 11:30am]
by Hugo Larochelle 27 Sep '07

27 Sep '07

This week's seminar (see http://www.iro.umontreal.ca/article.php3?id_article=107&lang=en) How Many Clusters? An Information-Theoretic Perspective by Susanna Still, Department of Information and Computer Sciences University of Hawaii Location: Pavillon Andre-Aisenstadt (UdeM), room 1409 Time: September 28th 2007, 11h30 Clustering provides a common means of identifying structure in complex data, and there is renewed interest in clustering as a tool for the analysis of large data sets in many fields. Anatural question is how many clusters are appropriate for the description of a given system. Traditional approaches to this problem are based on either a framework in which clusters of a particular shape are assumed as a model of the system or on a two-step procedure in which a clustering criterion determines the optimal assignments for a given number of clusters and a separate criterion measures the goodness of the classification to determine the number of clusters. In a statistical mechanics approach, clustering can be seen as a trade-off between energy- and entropy-like terms, with lower temperature driving the proliferation of clusters to provide a more detailed description of the data. For finite data sets, we expect that there is a limit to the meaningful structure that can be resolved and therefore a minimum temperature beyond which we will capture sampling noise. This suggests that correcting the clustering criterion for the bias that arises due to sampling errors will allow us to find a clustering solution at a temperature that is optimal in the sense that we capture maximal meaningful structure-without having to define an external criterion for the goodness or stability of the clustering. We show that in a general information-theoretic framework, the finite size of a data set determines an optimal temperature, and we introduce a method for finding the maximal number of clusters that can be resolved from the data in the hard clustering limit. _______________________________________________ Lisa_seminaires mailing list Lisa_seminaires(a)mercure.iro.umontreal.ca https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_seminaires

1 0

UdeM-McGill-MITACS machine learning seminar, PAA 1409 (UdeM), Fri Sept 28, 11:30am
by Hugo Larochelle 24 Sep '07

24 Sep '07

This week's seminar (see http://www.iro.umontreal.ca/article.php3?id_article=107&lang=en) How Many Clusters? An Information-Theoretic Perspective by Susanna Still, Department of Information and Computer Sciences University of Hawaii Location: Pavillon Andre-Aisenstadt (UdeM), room 1409 Time: September 28th 2007, 11h30 Clustering provides a common means of identifying structure in complex data, and there is renewed interest in clustering as a tool for the analysis of large data sets in many fields.Anatural question is how many clusters are appropriate for the description of a given system. Traditional approaches to this problem are based on either a framework in which clusters of a particular shape are assumed as a model of the system or on a two-step procedure in which a clustering criterion determines the optimal assignments for a given number of clusters and a separate criterion measures the goodness of the classification to determine the number of clusters. In a statistical mechanics approach, clustering can be seen as a trade-off between energy- and entropy-like terms, with lower temperature driving the proliferation of clusters to provide a more detailed description of the data. For finite data sets, we expect that there is a limit to the meaningful structure that can be resolved and therefore a minimum temperature beyond which we will capture sampling noise. This suggests that correcting the clustering criterion for the bias that arises due to sampling errors will allow us to find a clustering solution at a temperature that is optimal in the sense that we capture maximal meaningful structure-without having to define an external criterion for the goodness or stability of the clustering. We show that in a general information-theoretic framework, the finite size of a data set determines an optimal temperature, and we introduce a method for finding the maximal number of clusters that can be resolved from the data in the hard clustering limit.

1 0

ROOM CHANGE: UdeM-McGill-MITACS machine learning seminar, MC 437
by Hugo Larochelle 13 Sep '07

13 Sep '07

The UdeM-McGill-MITACS Machine Learning seminars are back! Note that this year, there will be seminars only every 2 weeks. The seminar will be in room 437, NOT 103! This week's seminar (see https://www.iro.umontreal.ca/article.php3? id_article=107&lang=en). Optimal Causal Inference by Susanna Still, Department of Information and Computer Sciences University of Hawaii Location: McConnell Engineering building (McGill), room 437 Time: September 14th 2007, 11h30 I will talk about how theory building can naturally distinguish between regularity and randomness. Starting from basic modeling principles I will argue for a general information-theoretic objective function that embodies a trade-off between a model’s complexity and its predictive power. The family of solutions derived from this principle corresponds to a hierarchy of models. At each level of complexity, those models achieve maximal predictive power, and in the limit of optimal prediction a process’ exact causal organization is identified. Examples show how theory building can profit from analyzing a process’ causal compressibility, which is reflected in the optimal models’ rate-distortion curve.

1 0

UdeM-McGill-MITACS machine learning seminar, MC 103 (McGill), Fry Sept 14, 11:30am
by Hugo Larochelle 10 Sep '07

10 Sep '07

The UdeM-McGill-MITACS Machine Learning seminars are back! Note that this year, there will be seminars only every 2 weeks. This week's seminar (see https://www.iro.umontreal.ca/article.php3? id_article=107&lang=en). Optimal Causal Inference by Susanna Still, Department of Information and Computer Sciences University of Hawaii Location: McConnell Engineering building (McGill), room 103 Time: September 14th 2007, 11h30 I will talk about how theory building can naturally distinguish between regularity and randomness. Starting from basic modeling principles I will argue for a general information-theoretic objective function that embodies a trade-off between a model’s complexity and its predictive power. The family of solutions derived from this principle corresponds to a hierarchy of models. At each level of complexity, those models achieve maximal predictive power, and in the limit of optimal prediction a process’ exact causal organization is identified. Examples show how theory building can profit from analyzing a process’ causal compressibility, which is reflected in the optimal models’ rate-distortion curve.

1 0