A numerical experiment on some gradient methods

Date

3-1991

Degree

Bachelor of Science in Applied Mathematics

College

College of Arts and Sciences (CAS)

Adviser/Committee Chair

Norberto R. Navarrete Jr.

Abstract

This paper studied the gradient methods for solving unconstrained minimization problem. These methods solve the problem by iteratively finding a descent direction using the gradient information and a step length along this direction. The steepest descent algorithm, Marguardt-Levenberg algorithm and the Davidon-Fletcher-Powell algorithm are some of the most commonly used technique under this class. Implementation of these algorithms were compared to the over-all efficiency and their performance per iteration. On the over-all performance level the Marguardt-Levenberg algorithm proved to be more efficient the other two. The iteration level comparison showed that a large amount of time was consumed by the Marguardt-Levenberg during the direction finding step while the other two spent most of their time in the step length evaluation.

Language

English

Location

UPLB Main Library Special Collections Section (USCS)

Call Number

Thesis

Document Type

Thesis

This document is currently not available here.

Share

COinS