Home | Search | Help  
Home Page Università di Genova

Seminar Details

Date 17-2-2010
Time 11:00
Room/Location DISI-Sala conferenze 3 piano
Title Bayesian Logistic Regression for Classification of Tabular Data
Speaker Dr. Dmitry Vetrov
Affiliation Lomonosov Moscow State University, CMC department
Link https://www.disi.unige.it/index.php?eventsandseminars/seminars
Abstract We consider classification problems with table-structured data, i.e. situations when objects are represented as tables of features. Important examples here are distances from an object to a set of support objects w.r.t. a set of features and image blocks/descriptors representation in image analysis. We extend the Relevance Vector Machine (RVM) framework to tabular data by coupling the regularization coefficients of rows and columns of features. We present two variants of this new gridRVM framework, based on the way in which the regularization coefficients of the rows and columns are combined. Appropriate variational optimization algorithms are derived for inference within this framework. The consequent reduction in the number of parameters from the product of the table's dimensions to the sum of its dimensions allows for better performance in the face of small training sets, resulting in improved resistance to overfitting problems, as well as providing better interpretation of results. In particular, the proposed technique allows to select simultaneously relevant objects and features in classification problems. Results are demonstrated on synthetic data-sets as well as on a modern and challenging visual identification benchmark.
Back to Seminars