Deep reinforcement learning methods for automated workflow construction in large scale open distributed systems

D'Arcy, Laura 2023. Deep reinforcement learning methods for automated workflow construction in large scale open distributed systems. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF (Laura D'Arcy, PhD, Thesis) - Accepted Post-Print Version Available under License Creative Commons Attribution No Derivatives. Download (3MB) | Preview
	PDF (Cardiff University Electronic Theses and Dissertations Form - Memorandum of Understanding) - Supplemental Material Restricted to Repository staff only Download (1MB)

Abstract

Large-scale distributed and decentralized systems often require access to multiple services, leading to the construction of complex workflows that can be difficult to design manually. This thesis proposes to use Deep Reinforcement Learning (DRL) techniques to create the optimal workflow without human intervention. The proposed hypothesis is based on using DRL algorithms combined with various styles of encoding such as Symbolic Vector Architecture and Knowledge Graph Embeddings, to handle larger and more complex systems. The approach utilizes both hierarchical and multi-task reinforcement learning. The benefit of using DRL in workflow construction is its ability to adapt to dynamic systems, where services are continuously added or removed, and systems change in quality. Our proposed approach can learn to adapt to changes in the system and find suitable alternatives.

Item Type:	Thesis (PhD)
Date Type:	Acceptance
Status:	Unpublished
Schools:	Computer Science & Informatics
Subjects:	Q Science > Q Science (General) T Technology > T Technology (General)
Funders:	DAIS ITA
Date of First Compliant Deposit:	13 March 2024
Date of Acceptance:	May 2023
Last Modified:	13 Mar 2024 16:46
URI:	https://orca.cardiff.ac.uk/id/eprint/167237

Actions (repository staff only)

Edit Item