I am a PostDoc at UCSB working with Professor Yao Qin.

I obtained my PhD from Brown University, working on Reinforcement Learning, advised by Professor George Konidaris. Prior to this I obtained a masters degree from University of Massachusetts, Amherst and had the privilege of working with Professor Phil Thomas. I graduated with a Bachelors of Technology in computer science at IIT Bombay. My research is focused on designing a geometric lense to understand and improve Deep Reinforcement Learning (RL).

High dimensional data has a low dimensional structure to it. For example, a picture of a face might have 2¹⁶ pixels but we can describe a face with far fewer variables: size of the nose, color of the skin, shape of the eyes etc. Similarly, a picture of robot might have numerous pixels but it can be described using a simple pose. This is the manifold hypothesis: high-dimensional data lies on (or close to) a low-dimensional manifold. Using this I answer three questions:

1. How do existing deep learning models utilise this low-dimensional structure?
2. How can we exploit this low-dimensional structure to train better models?
3. What is this structure for RL and how do we build agents that can utilise it?

I believe theory should move towards practice and practice should move towards theory in Deep RL for true progress. For details see list of my publications. I am working on deploying my theoretical framework to improve practice of deep RL for robotic control.