Jeudi 03 juillet 2014 de 14h00 à 15h30
Auditorium IRCICA, parc scientifique de la Haute Borne à Villeneuve d’Ascq
√ Abstract :
Online learning has become a standard tool in machine learning and large-scale data analysis. Learning is viewed as a repeated game between an adaptive agent and an ever-changing environment. Within this simple paradigm, one can model a variety of sequential decision tasks simply by specifying the interaction protocol and the resource constraints on the agent. In the talk we will describe algorithmic applications to specific learning scenarios (partial feedback, attribute-efficient, multitask, semi-supervised, and more). In order to provide a specific example of online learning, in the final part of the talk we will focus on the experts/bandits model and show a simple algorithmic setting that generalizes previous and seemingly different approaches.
√ Bio :
Nicolo Cesa-Bianchi is professor of Computer Science at the University of Milano, Italy. He is the former President of the Association for Computational Learning (2006-2009) and served as action editor for the Machine Learning Journal (2008-2012), for the IEEE Transactions on Information Theory (2011-2014), and for the Journal of Machine Learning Research (2009-2013). He is currently associate editor for the Journal of Information and Inference (2012-).
His main research interests include statistical learning theory, game-theoretic learning, and pattern analysis. He is co-author with Gabor Lugosi of the monograph « Prediction, Learning, and Games » (Cambridge University Press, 2006) and with Sébastien Bubeck of the monograph « Regret Analysis of Stochastic and Nonstochastic Multi-armed Bandit Problems » (Foundations and Trends in Machine Learning, 2012). He is recipient of a Google Research Award (2010) and a Xerox Foundation UAC award (2011).
