Developing foundations and principled, practical algorithms
for important modern learning paradigms. These include
interactive learning, distributed learning, transfer learning,
multi-task learning, learning representations, life-long
learning, and metalearning. My research formalizes and
explicitly addresses all constraints and important challenges of
these new settings, including statistical efficiency,
computational efficiency, noise tolerance, limited supervision
or interaction, privacy, low communication, and incentives.
Analyzing the overall behavior of complex systems in which
multiple agents with limited information are adapting their
behavior based on past experience, both in social and engineered
Computational aspects in game theory and economics.
Analysis of the algorithms beyond the worst case and more
generally identifying interesting and realistic models of
computation that provide a better alternative to traditional
worst-case models in a broad range of optimization problems.