Siddarth Chandrasekar

‎

I am a graduate student at the Computing Science department at the University of Alberta. I work on the intersection of representation learning and Reinforcement Learning with Marlos C. Machado. Currently I am working on learning reusable behavior bases for sample-efficient RL.

Previously, I was a Post-Baccalaureate Student at the RBCDSAI, working under the supervision of Prof. Ravindran. And before that, I spent some time at the SPIRE lab, IISc Bangalore, and the NLP Lab, IIT Patna. I completed my undergrad at IIIT-DM in Electronics and Communication Engineering.

Recent Works

The Laplacian Keyboard: Beyond the Linear Span

LK builds a library of reusable options from graph eigenvectors that a meta-policy dynamically combines to solve tasks beyond what any single option can express.

Reward-Aware Proto-Representations In Reinforcement Learning

We extend the SR by folding reward dynamics into the environment's structure, enabling reward-aware behavior useful for exploration, transfer, and planning.

Page updated

Google Sites

Report abuse