Specialization: Reinforcement Learning and Applications

My research is mainly focussed on reinforcement learning (RL) and stochastic optimization. I also work along with my students in the applications of machine learning methods in general to wireless communication, information retrieval and social networks.

Theory: On the theoretical front I have worked on problems from continuous state space RL, non-stationary RL. We have shown regret bounds for these problems. In the past we have worked on showing convergence of various RL algorithms using stochastic approximation theory. In stochastic optimization we have tired coupling quasi-Newton, accelerated methods with standard stochastic gradient search methods. We have showed convergence in these settings. We also tested them in simple experiments.

Applications: We have looked at the application to the areas of wireless communication, information retrieval and social networks. More particularly we have focussed on developing non-stationary bandit algorithms for beam-forming in 5G wireless networks. We are working on reinforcement-based hybrid recommendation systems. In the broad area of social networks, we have focussed on influence maximization and link prediction in dynamic networks using a variety of ML algorithms.