Invited Talk by Dr. Chandrashekar on Planning Under Uncertainty
Exact methods to solve Markov Decision Processes (MDPs) are ineffective in practice either because there are a large number of states or because the model of the MDP is not known. Approximate Dynamic Programming (ADP) algorithms are approximate solution methods for MDPs with large number of states. Reinforcement Learning (RL) algorithms are sample trajectory based solution methods for MDPs. The talk focuses on conditions that guarantee performance of ADP algorithms and stability of RL algorithms.
Chandrashekar L, completed his PhD at the Department of Computer Science and Automation, Indian Institute of Science, Bangalore. He was advised by Prof. Shalabh Bhatnagar and his PhD research deals with developing Approximate Dynamic Programming (ADP) and Reinforcement Learning (RL) algorithms with provable performance and guaranteed convergence. He is also interested in stochastic optimization, machine learning and related areas. He obtained the 1st rank in GATE in 2008, and subsequently completed his Masters in Systems Science and Automation from Indian Institute of Science, Bangalore (2010) after a Bachelors in Instrumentation and Control Engineering from National Institute of Technology, Trichy (2005). During the period from 2005 to 2008, he was an Analog design engineer and was involved in building high speed analog-to-digital converters.