Harish Kamath

Georgia Institute of Technology

Atlanta, GA

Hey! I’m an undergraduate in my 4^th and final year at Georgia Tech. I am pursuing a double major in Computer Science and Mathematics, with a focus in artificial intelligence and systems architecture. I am particularly passionate about research and startups in the tech space!

Research Interests:

Generalization theory in deep learning - performance of models across changing distributions, alternatives to ERM, and the roles of sparsity and unsupervised feature learning in robust models. I find that generalization in reinforcement learning is especially interesting, due to its applications in model-based RL, sim2real, and exploration strategies.

Other things about me:

I enjoy playing & following basketball (go Hawks!), going on runs & hikes, and chess.

news

Aug 28, 2020	I completed my ML Engineering internship at Ripcord
Aug 24, 2020	I presented my RLOSFest project to the MSFT New York Research Lab
Jul 18, 2020	We presented our paper on model-advantage at two ICML Workshops! (1 2)

selected publications

Workshop@ICML

Bridging Worlds in Reinforcement Learning with Model-Advantage

Modhe, N., Kamath, H., Kalyan, A., and Batra, D.

2020

Abs PDF

Despite the breakthroughs achieved by Reinforcement Learning (RL) in recent years, RL agents often fail to perform well in unseen environments. This inability to generalize to new environments prevents their deployment in the real world. To help measure this gap in performance, we introduce model-advantage - a quantity similar to the well-known (policy) advantage function. First, we show relationships between the proposed modeladvantage and generalization in RL — using which we provide guarantees on the gap in performance of an agent in new environments. Further, we conduct toy experiments to show that even a sub-optimal policy (learnt with minimal interactions with the target environment) can help predict if a training environment (say, a simulator) helps learn policies that generalize. We then show connections with Model Based RL.