Rishabh Agarwal
Reinforcement Learner at McGill University
Canada
Summary
👤
Senior🎓
Top SchoolRishabh Agarwal is a reinforcement learning researcher and engineer with 11 years of experience bridging deep RL research and large-scale ML systems, currently based in Canada and holding roles at Periodic Labs and McGill University. He has driven RL innovations at Google Brain and DeepMind—contributing to influential projects like RL Unplugged and human-level Atari efficiency—and helped scale RL tooling in prominent open-source repos such as Google Research, Acme, and Dopamine. His work spans algorithm design, distributional value functions, memory- and GPU-efficient implementations, and practical fixes that improved reproducibility and experiment configuration for the community. Rishabh has also led RL-for-LLM research and distillation efforts at Meta’s Superintelligence Labs and Google DeepMind, blending RL with large language model reasoning. He holds a PhD-level affiliation with Mila and a CS degree from IIT Bombay, reflecting a strong academic foundation paired with hands-on system engineering. An underappreciated thread in his profile is consistent low-level robustness work—debugging distributions, installation issues, and logging—that materially improved research throughput for others.
10 years of coding experience
7 years of employment as a software developer
Doctor of Philosophy - PhD Artificial Intelligence, Doctor of Philosophy - PhD Artificial Intelligence at Mila - Quebec Artificial Intelligence InstituteIndian Institute of Technology Bombay
Github Skills (12)
machine-learning10
agent10
tensorflow10
jax10
python10
reinforcement-learning10
tensorflow-probability9
ai9
dqn9
deep-learning8
data-augmentation8
ml7
Programming languages (7)
Github contributions (5)
google-research/google-research
Feb 2019 - Aug 2022
Google Research
Role in this project:
ML Engineer
ML EngineerContributions:29 commits, 2 PRs, 9 comments in 3 years 6 months
Contributions summary:Rishabh contributed code for a reinforcement learning agent within the Google Research repository. This involved adding an instruction-following task for MeRL, suggesting involvement in meta-reward learning. The commits demonstrate an understanding of model implementation and potentially optimization strategies for reinforcement learning problems.
googlemachine-learningai
google/dopamine
May 2019 - Nov 2022
Dopamine is a research framework for fast prototyping of reinforcement learning algorithms.
Role in this project:ML Engineer
ML EngineerContributions:25 commits, 1 comment in 3 years 6 months
Contributions summary:Rishabh made several contributions focused on improving the reinforcement learning framework. They fixed a bug in the log reading process, allowed for more efficient GPU memory usage, and added logging statements within the DQN agent. Furthermore, the user added support for distributional value functions (vmin) in Rainbow agents and fixed Rainbow to correctly compute gradients using loss weighted by inverse priorities. They also moved input normalization into a preprocessing function. Finally, the user added code and configurations for running Atari 100k experiments, including data augmentation techniques, and fixed no-ops logic.
reinforcement-learning-algorithmsfast-prototypinggooglereinforcement-learningprototyping
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.