Design and evaluation of a hybrid multi-task learning model for optimizing deep reinforcement learning agents

Driven by recent technological advancements within the artificial intelligence domain, deep learning has emerged as a promising representation learning technique. This in turn has given rise to the evolution of deep reinforcement learning that combines deep learning with reinforcement learning methods. Subsequently, performance optimization achieved by reinforcement learning intelligent agents designed with model-free based approaches were predominantly limited to systems with reinforcement learning algorithms learning single task. Such a model was found to be quite data inefficient, whenever agents needed to interact with more complex, rich data environments. This thesis introduces a hybrid multi-task learning-oriented approach for optimization of deep reinforcement learning agents operating within different but semantically similar environments with related tasks. Empirical results obtained with OpenAI Gym library-based Atari 2600 video gaming environment demonstrate that the proposed hybrid multi-task learning model is successful in addressing key challenges associated with the performance optimization of deep reinforcement learning agents.

Name thesis:degree_name

Master of Applied Science (MASc)

Discipline thesis:degree_discipline

Electrical and Computer Engineering

Grantor

University of Ontario Institute of Technology

Year dc:date.issued

2021

Author dc:creator

• Vithayathil Varghese, Nelson

Advisor dc:contributor.advisor

• Mahmoud, Qusay H.

• Deep reinforcement learning
• Neural networks
• Deep learning
• Multi-task learning
• Actor-critic

Language dc:language.iso

en

Handle dc:identifier.uri

https://hdl.handle.net/10155/1303

OAI identifier oai:identifier

oai:ontariotechu.scholaris.ca:10155/1303