Amin Rakhsha

About Me

Hi! My name is Amin Rakhsha and I am a PhD student in Computer Science at the University of Toronto under the supervision of Prof. Amir-massoud Farahmand.

I am broadly interested in designing and understanding agents that can learn about their environment through active interactions, and perform a variety of tasks in the environment.

My research focuses on developing robust and accelerated planning algorithms in reinforcement learning. I investigate how agents can effectively utilize a model of the environment while mitigating the impact of model inaccuracy. Additionally, I explore techniques like temporal abstraction to accelerate planning in tasks with long time horizons.

I am also interested in sample complexity analysis and exploration in reinforcement learning.

For more details, see my CV.

Recent News

• [September 2022] Our paper "Operator Splitting Value Iteration" was accepted at NeurIPS 2022!
• [May 2022] I was honored to receive the Borealis AI Fellowship.
• [December 2020] I was honored to receive the Computer Science 50th Anniversary Graduate Scholarship from department of computer science at University of Toronto.
• [December 2020] Our journal paper "Policy Teaching in Reinforcement Learning via Environment Poisoning Attacks" was accepted at JMLR.
• [June 2020] Our paper "Policy Teaching via Environment Poisoning: Training-time Adversarial Attacks against Reinforcement Learning" was accepted at ICML 2020.
• [April 2020] I will join the University of Toronto as a PhD student in Computer Science!

Publications

Operator Splitting Value Iteration

Reward poisoning in reinforcement learning: Attacks against unknown learners in unknown environments [arxiv]

Policy Teaching in Reinforcement Learning via Environment Poisoning Attacks [arxiv]

Policy Teaching via Environment Poisoning: Training-time Adversarial Attacks against Reinforcement Learning [arxiv]

* Equal Contribution

Research Experience

Max Planck Institute for Software Systems (MPI-SWS)

• Formulated the problem of adversarial attacks in reinforcement learning. 
• Inspected the problem of poisoning rewards in online and offline RL settings.

Chinese University of Hong Kong (CUHK)

• Worked on improving the optimization algorithm used for distributionally robust logistic regression under supervision of Prof. Anthony Man-Cho So.
• Analyzed randomness extraction from generalized Santha-Vazirani sources with infinite dice under supervision of Prof. Andrej Bogdanov.

Education

University of Toronto

Sharif University of Technology

Young Scholars Club

Allameh Helli High School

Teaching

Teaching Assistant

• CSC384: Introduction to Artificial Inteligence - Winter 2021 - Fall 2022
• CE354: Design of Algorithms - Fall 2019
• CE282: Linear Algebra - Fall 2018
• CE254: Data Structures and Algorithms - Fall 2018
• CE181: Probability and Statistics - Fall 2017