Developing new fingerprint identification system

Dissertant

Abd al-Husayn, Alya

University

University of Technology

Faculty

-

Department

Computer Sciences Department

University Country

Iraq

Degree

Master

Degree Date

2006

English Abstract

Branch prediction unit is an important part of modem processor architecture.

Its responsibility is to predict whether branches will be taken or not taken before they are actually executed.

In contrast, most branch prediction research focuses or ; Dynamic Branch Prediction (at run-time) as Two-Level Branch Prediction techniques, a very specific solution to the branch prediction problem.

That is a commonly used history table of two-bit saturating counters.

In this thesis we consider the application of artificial intelligence learning methods, particularly the use of Neural Networks to the branch prediction problem as an alternative approach.

This work presents a new method for branch prediction.

The key idea is to use the standard single and multilayer neural networks.

We use single layer perceptron based on perceptron learning algorithm and multilayer perceptron based on standard learning algorithm (Back propagation).

The designed neural net has been tested for operating branches, particularly conditional branch instructions, when they are being executed in microprocessor for different applications.

A comparative analysis and studies have been earned out with the other known prediction techniques.

The achieved results of our study show very high prediction accuracy rates.

The prediction accuracy rates arc achieved by the non-adaptive neural predictors and conventional predictors, the neural predictors are better than conventional predictors, but in adaptive the neural predictors is comparable to conventional two-level predictors with the same size of input.

But when regarding the same hardware budget the neural predictors are the best, but they have taken more time consumed for computation branch prediction than conventional predictors.Branch prediction unit is an important part of modem processor architecture.

Its responsibility is to predict whether branches will be taken or not taken before they are actually executed.

In contrast, most branch prediction research focuses or ; Dynamic Branch Prediction (at run-time) as Two-Level Branch Prediction techniques, a very specific solution to the branch prediction problem.

That is a commonly used history table of two-bit saturating counters.

In this thesis we consider the application of artificial intelligence learning methods, particularly the use of Neural Networks to the branch prediction problem as an alternative approach.

This work presents a new method for branch prediction.

The key idea is to use the standard single and multilayer neural networks.

We use single layer perceptron based on perceptron learning algorithm and multilayer perceptron based on standard learning algorithm (Back propagation).

The designed neural net has been tested for operating branches, particularly conditional branch instructions, when they are being executed in microprocessor for different applications.

A comparative analysis and studies have been earned out with the other known prediction techniques.

The achieved results of our study show very high prediction accuracy rates.

The prediction accuracy rates arc achieved by the non-adaptive neural predictors and conventional predictors, the neural predictors are better than conventional predictors, but in adaptive the neural predictors is comparable to conventional two-level predictors with the same size of input.

But when regarding the same hardware budget the neural predictors are the best, but they have taken more time consumed for computation branch prediction than conventional predictors.

Main Subjects

Information Technology and Computer Science

Topics

• Data processing

• APA
• MLA
• AMA

Language

English

Data Type

Arab Theses

Record ID

BIM-305769