| Start Date: | 4/11/2014 | Start Time: | 1:00 PM |
| End Date: | 4/11/2014 | End Time: | 3:00 PM |
|
Event Description
Title: Meta-Subset Selection for Detection of Variable Importance in the Microbiome
Advisor: Dr. Gail Rosen
Date: Friday, April 11, 2014
Time: 1:00 p.m.
Location: Biomed Seminar Room 709, 7th Floor, Bossone Research Enterprise Center
Abstract
Subset selection is a widely studied research problem in machine learning because the complexity of most computational intelligence algorithms, as well as the amount of data needed to obtain a robust model with such an algorithm, increases rapidly with the number of features. Despite several well-established subset selection algorithms that have been developed over the years, the development of a computationally-efficient sequential framework that can determine the most relevant features for a user-defined objective function has been understudied. Some wrapper and embedded methods can select the most important features with little to no prior information; however, such methods must also learn the classifier parameters, which can be computationally burdensome or intractable for incremental learning.
This research proposal addresses the issue of identifying feature importance using filter-based subset selection algorithms. The first aim of the proposal addresses the detection of feature importance via a Neyman-Pearson hypothesis test used in conjunction with a generic subset selection algorithm. The Neyman-Pearson approach considers the entire feature set to detect importance, which could limit the spectrum of its applicability for large databases. Therefore, in the second aim we propose to use sequential learning algorithms, which only consider a subset of the features at each learning round, to detect variable importance with a generic subset selection algorithm. We consider the proposed subset selection algorithms to infer the importance of bacteria in microbial communities, which has broad impacts in the study of human health and the environment. |
|
Location:
Biomed Seminar Room 709, 7th Floor, Bossone Research Enterprise Center |
Audience:
Current StudentsFacultyStaff |
|
