Majority Problems in Distributed Systems and Clustering in Structured Graphs



Hamilton, DD
(2017) Majority Problems in Distributed Systems and Clustering in Structured Graphs. PhD thesis, University of Liverpool.

[img] Text
200677509_June2017.pdf - Unspecified

Download (3MB)

Abstract

This thesis focuses on the study of various algorithms for Distributed Computing and Machine Learning research areas. More precisely, the work within contains research into various communication protocols in different settings of Distributed Computing, accompanied by relevant analysis on protocol performance in time and space. These protocols are designed to operate in analogous environments using different models for communication, primarily population protocol and random walk variants. In our settings we aim to use as minimal memory as possible, achieving light weight protocols that are powerful in their capabilities and randomized as well as deterministic in nature. We also propose a novel technique of verification which enables multi-step protocols to work in synergy. These protocols generally never terminate, but converge and are difficult to disseminate results throughout the network to be used in dependent processes. With the verification technique proposed, protocols can become adaptive and stacked into a chain of dependent processes. We also provide experimental analysis of a subarea of Machine Learning, unsupervised clustering algorithms. Gaining inspiration from the agglomerative nature and techniques defined in classical hierarchical clustering as well as the Phylogenetic tree building methods, we provide a comprehensive study and evaluation of new method to agglomeratively combine `similar' data into clusters based on the general consensus of taxonomy and evaluation of clustering mechanisms.

Item Type: Thesis (PhD)
Divisions: Faculty of Science and Engineering > School of Electrical Engineering, Electronics and Computer Science
Depositing User: Symplectic Admin
Date Deposited: 23 Aug 2017 10:11
Last Modified: 19 Jan 2023 07:02
DOI: 10.17638/03008152
Supervisors:
URI: https://livrepository.liverpool.ac.uk/id/eprint/3008152