Varun Giridhar
Varun Giridhar is a Master's student in Computer Science at Georgia Institute of Technology, where he is advised by Animesh Garg in the PAIR research group. His research focuses on reinforcement learning, world models, and gradient-based optimization for robotics. He is best known for his work on PWM (Policy Learning with Multi-Task World Models), which was presented at ICLR 2025, and his research on dynamics decomposition for multi-task reinforcement learning.
“The core breakthrough of this paper is exploiting the asymmetry between behavior cloning (BC) and Q-learning: BC can only be trained on successful demonstrations, while an off-policy Q-function can be trained on *any* rollout, including failures.”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.