MSc Essay Presentation - Haley Li

Name: Haley Li

Date: June 18

Time: 11:00 AM – 12:00 PM

Location: ICICS 306

Supervisor: Mathias Lécuyer and Thomas Pasquier

Title: Reinforcement Learning for Distributed Systems Policies: A Literature Review into Current Approaches and Challenges Faced

Abstract: In the field of distributed systems, heuristic policies are typically used to make decisions within the system under the objective of improving performance. However, designing good heuristics often requires deep understanding of the complex intricacies within the system. Reinforcement learning (RL) grants a framework to learn such policies without needing as much system knowledge by learning through interaction. However, the process of integrating RL involves challenges that make their adoption in distributed systems difficult. RL agents require lots of potentially risky exploration decisions due to large state spaces, they may take lots of time to train, and they may perform poorly in training or unpredictably overall. We perform a literature review that investigates the challenges faced by current research in adopting RL for distributed systems policies and their propositions in mitigating those challenges.