Place: Large Lecture Room – CVC
Affiliation: Chaire de recherche du CRSNG Dept informatique Universite de Sherbrooke Quebec, Canada
The democratization of information and communication technologies is making available huge quantities of data. Using this data in efficient ways will help to improve the activity of many sectors in different areas. In this regard, during the last few decades, methodologies, models, algorithms, and systems of machine learning were revisited; however, additional efforts are required to propose effective solutions to some open problems; among them — scalability, dimensionality, feature selection, and updating. During the past few years, my collaborators and I have proposed several machine learning algorithms to approach these problems in the case of both finite and infinite mixture models, as well as their use in real-world applications. This talk will focus on the learning of statistical models in the case of mixture of pdfs; specifically, the discriminative and generative learning, non-Gaussian data modeling, model selection, feature in the case of high dimensional space, and updating of mixture models. I will also illustrate the developed algorithms in the context of the recommendation of images.