Our group conducts research on fundamental problems in machine learning, convex optimization and game theory.

Online Learning

Online learning is a mathematical optimization framework that captures problems that take the form of playing a repeated game against an adversary. Examples for problems of interest that fall into the online learning framework include: online portfolio selection, designing recommendation systems, online navigation, time series analysis, statistical learning and more. The quality of online learning algorithms is measured by a quantity known as regret which is the difference between the performance of the online algorithm and that of an optimal offline benchmark.

Our group develops state-of-the-art algorithms for various online learning problems with the emphasis on computational efficiency and optimal prediction/regret bounds. These algorithms range from general-case online learning to domain-specific. Recent examples include  Adversarial MDP LearningOnline Matrix PredicationTime Series PredictionLinear Regression with Limited FeedbackUniversal Filtering

Efficient Optimization for Machine Learning

In many modern data-intensive applications, computational efficiency is a major concern any super-linear time operation is intractable. A major research effort of our group is dedicated to the development of highly-efficient, linear or sublinear-time algorithms for very-large scale optimization problems, usually in the context of machine learning. Recent directions include:


The research leading to these results has received funding from the People Programme (Marie Curie Actions) of the European Union’s Seventh Framework Programme FP7/2007-2013/ under REA grant agreement n° [276840].