Latest News

  • September 2024. “Efficient Reinforcement Learning by Discovering Neural Pathways” got accepted in NeurIPS 2024! 🎉
  • March 2024. I’m joining Microsoft Research, Montreal as part-time research intern. I will be working on parameter efficient fine-tuning (PEFT) with the goal of building a mixture of experts (MoE) for LLMs.

I am a visiting student-researcher at Microsoft Research, Montreal and doing a Ph.D. in Computer Science at McGill University and Mila Quebec AI Institute working with Dr. Doina Precup.

My research focus is on “Improving the Learning Capacity and Parameter Efficient training for RL and LLMs”. I am working on efficiently using neural networks where we take inspiration from the human brain, using multiple specialized pathways through a single network, with each pathway focusing on a single task. This is an alternate way to “routing” and a mixture of expert structures that can be added to LLM.

I’m interested in the Mixture of experts (MoE), Parameter-efficient finetuning (peft) in LLM, Preference fine-tuning using Reinforcement Learning (RL), LLM alignment, Improving mergability of a mixture of experts.

Prior I did applied research internships at Microsoft Research, New York (summer 2023, host: John Lanford, Alex Lamb), Ubisoft La Forge, Montreal (2021-2022, host: Joshua Romoff), Mila Quebec AI Institute (2019, host: Doina Precup)

I completed Master’s in Electrical and Computer Engineering at McGill University. My master’s research was on “Adversarial Inverse Reinforcement Learning” under the supervision of Dr. Aditya Mahajan at Centre for Intelligent Machine (CIM).

News

  • May, 2023- Aug, 2023. I worked at Microsoft Research, New York as Research Intern, on an Appiled RL project with John Langford and Alex Lamb.
  • October, 2021. Two papers got accepted in Offline Reinforcement Learning Workshop, NeurIPS 2021
    • “Importance of Empirical Sample Complexity Analysis for Offline Reinforcement Learning” - Paper
    • “Single-Shot Pruning for Offline Reinforcement Learning” - Paper
  • Sep, 2021- Aug 2022. I have worked at Ubisoft, Montreal with Joshua Romoff as Research Intern.
  • June, 2020. “Off-Policy Adversarial Inverse Reinforcement Learning” got accepted in Lifelong Learning workshop, ICML 2020. Paper, Code,Talk
  • January, 2020. I have started my Ph.D. at McGill University.
  • June, 2019. I joined Mila as Research Intern.
  • June, 2019. “Doubly Robust Estimators in Off-Policy Actor-Critic Algorithms” got accepted for spotlight presentation at RLDM 2019
  • January, 2018. I started my Master’s at McGill University.

Research Interest

  • Reinforcement Learning: Improving LLM, Imitation Learning, Offline Reinforcement Learning, Multitask Learning, Representation learning
  • Large Language Model: Mixture of experts (MoE), Improving mergability of a mixture of experts, Parameter-efficient finetuning (peft), Preference fine-tuning, LLM alignment