A Classification and Comparative Study of Existing Fast Reinforcement Learning Techniques

Deep reinforcement learning has achieved great successes in recent years. One of the primary challenges faced by such methods is the high cost involved in training a system that demonstrates the desired competency and performance. This project aims to study and compare the available techniques for improving the training efficiency and effectiveness of reinforcement learning and establish a method of integrating such techniques to the existing models. The proposed project consists of conducting a review on fast reinforcement learning techniques, categorizing the key recent innovations to be explored, run an empirical comparative study between the corresponding techniques and create an internal library allowing to integrate these different techniques for future investigation and implementation.

Faculty Supervisor:

Ioannis Mitliagkas

Student:

Partner:

Solid State of Mind Inc

Discipline:

Engineering

Sector:

Professional, scientific and technical services

University:

Université de Montréal

Program: