Talk 1: Policy Optimization in Reinforcement Learning
Abstract: The main goal of reinforcement learning is to come up with an optimal policy through which an agent can seamlessly interact with the uncertain and often semi-observable external world (environment). Applying policy gradient methods to search for the optimal policy has been found increasingly useful and widely applicable in large scale problems across a wide variety of domains. In this talk, we aim to present the state-of-the-art policy gradient methods and show the efficacy of various policy optimization techniques in the traditional reinforcement learning setup.
Talk 2: Learning to become Experts in Playing Games
Abstract: From early days since the inception of artificial intelligence, significant breakthroughs are being made to master the art of playing various strategic games and outperform humans. In one hand, the functional rules of the game are simple enough to learn by the game-playing agents, but achieving master level performance involves understanding the intricate decision-making abilities against intelligent and unforeseen opponent (sometimes expert humans). In this talk, we aim to present the dominance of artificial intelligence methods and reinforcement learning algorithms in game-playing and show how these intelligent techniques are achieving expert-level performance by challenging (and many-a-times outperforming) the professional human players.
Talk 3: Formal Methods for Power Intent Verification
Abstract: With increasing functionalities of modern low-power integrated circuits, sophisticated power management strategies are deployed to extend the battery life. These strategies involve shutting down parts of the circuit that are not in use, and operating different parts of the circuit in different voltage-frequency combinations thereby balancing power consumption and performance. The power management strategy is decided up front by the architects of the chip, and must be proven to be faithfully implemented in the digital as well as analog (mixed-signal) design counterparts. In this talk, we aim to present the strategies for architectural power intent verification of mixed-signal domains in integrated circuits at the digital-analog boundary is a part of their power management requirement.
Qualifications: Doctor of Philosophy (Ph.D.), IIT Kharagpur, 2015
Affiliation: Indian Institute of Technology Kharagpur (Paschim Medinipur, West Bengal - 721302)
Position: Assistant Professor
Email: [email protected]