Dhruv Sreenivas

I am a PhD student at Mila, Université de Montréal, where I am advised by Pablo Samuel Castro. I completed my BS and MS at Cornell University, where I was advised by Wen Sun.

CV  /  LinkedIn  /  Email  /  GitHub

I'm primarily interested in (deep) reinforcement learning, imitation learning, and deep generative models.

We improve generalization in goal-conditioned behavior cloning policies via sampling goals from a geometric horizon model, allowing for experience across trajectories to be stitched together to provide useful goals for conditioning.

We find that making the critic "optimistic" in actor-critic learning (e.g. by doing Q value reduction differently) significantly improves downstream performance of much smaller actors across a variety of control domains.

By viewing off-policy adversarial imitation learning through the framework of gradient boosting, we develop a novel, theoretically principled algorithm that outperforms state-of-the-art methods.

We leverage advances in deep learning to propose a universal multimodal deconfounder for causal inference, with applications in genome-wide association studies (GWAS).

We leverage offline data with partial coverage over the expert state distribution to mitigate the covariate shift problem in imitation learning.

• Spring 2023: Graduate TA for CS 6789: Foundations of Reinforcement Learning (PhD)
• Fall 2022: Graduate TA for CS 6756: Learning for Robot Decision Making (PhD)
• Spring 2022: Graduate TA for CS 4789: Introduction to Reinforcement Learning
• Fall 2021: Graduate TA for CS 2110: OOP & Data Structures

• Reviewer: NeurIPS (2023, 2025), ICLR (2024, 2025), RLC (2025)

Website template is from Jon Barron's website.