Ethan Criss

Abstract

Starfish Search is a swarm optimization algorithm that operates in the same vein as

Particle Swarm Optimization and the Firefly Algorithm. This search algorithm attempts to find global optimal solutions to optimization problems by dispersing agents into the search space. Each agent consists of many nodes that represent candidate solutions to the problem being solved. Agent's nodes are formatted in a parent-child hierarchy, similar to tree structures, which facilitates information passing to a root node. With this structure, it becomes possible to determine the likely direction in which an optimal lies. By using a form of linear regression, the fitness values and positions of each node in an agent are used to evaluate a vector, known as the Local Gradient. This vector points along the slope of the search space, and its magnitude represents the steepness of this slope. In this way, an agent has an understanding of the local area and can make intelligent decisions about which direction to search for additional candidate solutions. With this additional information, agents also have the ability to execute behaviors based on the type of topology encountered. These behaviors can be specifically tailored to individual problems and situations to help agents correctly solve the problem.

Starfish Search has been applied to problems such as, search space optimization, k nearest neighbors classification, and k means clustering. By tailoring fitness functions and behavior execution, evidence has been gathered to support the algorithms use over traditional techniques. This paper dives into the details of the algorithm's implementation, calculations, and behaviors as well as explain the tests and evidence gathered to support the use of Starfish Search.

Library of Congress Subject Headings

Swarm intelligence; Computer algorithms--Testing; Database searching; Data mining

Publication Date

11-21-2013

Document Type

Thesis

Student Type

Graduate

Degree Name

Computer Science (MS)

Department, Program, or Center

Computer Science (GCCIS)

Advisor

Roger S. Gaborski

Advisor/Committee Member

Peter G. Anderson

Advisor/Committee Member

Joe Geigel

Comments

Physical copy available from RIT's Wallace Library at Q337.3 .C74 2013

Campus

RIT – Main Campus

Download

Share

COinS