Abstract

The rapid advancement in the capacity and reliability of data storage technology has allowed for the retention of virtually limitless quantity and detail of digital information. Massive information databases are becoming more and more widespread among governmental, educational, scientific, and commercial organizations. By segregating this data into carefully defined input (e.g.: images) and output (e.g.: classification labels) sets, a classification algorithm can be used develop an internal expert model of the data by employing a specialized training algorithm. A properly trained classifier is capable of predicting the output for future input data from the same input domain that it was trained on. Two popular classifiers are Neural Networks and Support Vector Machines. Both, as with most accurate classifiers, require massive computational resources to carry out the training step and can take months to complete when dealing with extremely large data sets. In most cases, utilizing larger training improves the final accuracy of the trained classifier. However, access to the kinds of computational resources required to do so is expensive and out of reach of private or under funded institutions. The Cell Broadband Engine (CBE), introduced by Sony, Toshiba, and IBM has recently been introduced into the market. The current most inexpensive iteration is available in the Sony Playstation 3 ® computer entertainment system. The CBE is a novel multi-core architecture which features many hardware enhancements designed to accelerate the processing of massive amounts of data. These characteristics and the cheap and widespread availability of this technology make the Cell a prime candidate for the task of training classifiers. In this work, the feasibility of the Cell processor in the use of training Neural Networks and Support Vector Machines was explored. In the Neural Network family of classifiers, the fully connected Multilayer Perceptron and Convolution Network were implemented. In the Support Vector Machine family, a Working Set technique known as the Gradient Projection-based Decomposition Technique, as well as the Cascade SVM were implemented.

Library of Congress Subject Headings

Information storage and retrieval systems--Data processing; Classification--Data processing; Neural networks (Computer science); Machine learning; Computer algorithms

Publication Date

8-1-2008

Document Type

Thesis

Department, Program, or Center

Computer Engineering (KGCOE)

Advisor

Cockburn, Juan

Advisor/Committee Member

Melton, Roy

Comments

Note: imported from RIT’s Digital Media Library running on DSpace to RIT Scholar Works. Physical copy available through RIT's The Wallace Library at: QA76.9.D37 W94 2008

Campus

RIT – Main Campus

Download

Share

COinS