Arushi Jain

I am second year Computer Science (Artificial Intelligence) Ph.D. candidate with Doina Precup and Pierre-Luc Bacon in Reasoning and Learning Lab (RLLab) at McGill University and Mila, Montreal , Canada.

I am currently hosted by Alessandro Lazaric for internship at Meta AI Research Lab (FAIR) in Paris. I am working on learning a self-supervised framework that ties exploration and representation together in reward-free setting.

I did my Masters at McGill University and Mila, with Doina Precup. Before starting Masters, I was a Research Fellow at Microsoft Research (MSR), India in ML team, where I was advised by Sundararajan Sellamanickam . I did my Bachelors in Computer Science at IIIT-Delhi, India. There, I worked in Image Analysis and Biometrics (IAB) lab, where I was advised by Mayank Vatsa and Richa Singh.

Broadly my research interests span reinforcement learning, safety in AI and exploration.

Email / CV / Google Scholar / Twitter / Github

Highlights and News

03/10/2022 Interning with Alessandro Lazaric in Reinforcement Learning team at FAIR, Paris. I am working on learning better exploration and representations in unsupervised reward-free setting to produce optimal policy in zero-shot manner.

04/07/2022 Interning with Personalization team at Amazon, working on RL for Recommender Systems starting July 2022.

01/07/2022 Thrilled to be among the 10 recipients who are awarded "Borealis AI 2022 Fellowship". Press coverage by McGill University.

20/05/2022 "Towards Painless Policy Optimization for Constrained MDPs" paper accepted at UAI, 2022.

18/03/2022 "Robust Constrained MDPs" paper accepted at RLDM, 2022.

24/01/2022 Accepted to AGI Safety Fundamentals course by Effective Altruism Cambridge (EAC). I am taking part in Tech Alignment track.

Research

My research focuses on learning safe reinforcement learning (RL) algorithms grounded in theory which can be extended to the real-world applications. I am also interested in off-policy RL and Constrained MDPs (CMDPs).

	Towards Painless Policy Optimization for Constrained MDPs Arushi Jain, Sharan Vaswani, Reza Babanezhad, Csaba Szepesvari, Doina Precup UAI and RLDM, 2022 paper / short RLDM paper / code / RLDM poster
	Variance Penalized On-Policy and Off-Policy Actor-Critic Arushi Jain, Gandharv Patil, Ayush Jain, Khimya Khetarpal, Doina Precup AAAI, 2021 paper / code / talk / slides / poster
	Safe Option-Critic: Learning Safety in the Option-Critic Architecture Arushi Jain, Khimya Khetarpal, Doina Precup Knowledge Engineering Review (KER) Journal, 2021. (Cambridge University Press Journal) Adaptive Learning Agents (ALA) Workshop, ICML, 2018. paper / code / slides / poster
	Safety using Constraint Variance in Policy-Gradient Methods Arushi Jain Master's thesis, McGill University, March 2020. paper
	Safe Actor-Critic Arushi Jain, Ayush Jain, Doina Precup Safety, Risk and Uncertainty in RL Workshop, UAI, 2018. Women in ML (WiML) Workshop, NeurIPS* , 2018. paper / / slides / poster
	Safe Hierarchical Policy Optimization using Constrained Return Variance in Options Arushi Jain, Doina Precup RLDM*, 2019. paper
	Learning Options using Constrained Return Variance Arushi Jain, Doina Precup Safety and Robustness in Decision Making Workshop , NeurIPS*, 2019. paper / poster
	Safe Policy Learning with Constrained Return Variance Arushi Jain* Graduate Student AI Symposium, Canadian AI Conference, 2019. Proceeding published in LNAI Series by Springer Verlag. paper / talk

Reviewer

Reviewer, DARL, ICML Workshop ('22)

Reviewer, AISTATS ('22)

Reviewer, ICLR ('22), ML Evaluation Standards Workshop

Reviewer, NeurIPS ('18), WiML Workshop

Source code and style from Jon Barron's website.