Classifying Handwritten Digits with the Artificial Neural Network and Genetic Algorithm in a Distributed Environment

From CSI 702

Jump to: navigation, search

Contents

  1. Classifying Handwritten Digits with the Artificial Neural Network and Genetic Algorithm in a Distributed Environment
    1. Abstract
    2. Speedup Summary
    3. Image Gallery

1 Classifying Handwritten Digits with the Artificial Neural Network and Genetic Algorithm in a Distributed Environment

1.1 Abstract

This study explores the integration of the genetic algorithm and artificial neural network in performing classification of handwritten digits in a distributed environment. Message Passing Interface (MPI) is used in the modeling of the environment in which many configurations of the artificial neural network are initiated and OpenMP is used to increase the runtime performance of the artificial neural network calculations. Although implementations of the serialized backpropagation algorithm converge on a solution more rapidly than the distributed implementation, the backpropagation algorithm has a tendency to converge into local minima, requiring additional tuning parameters to avoid such behavior. A superposition of both backpropagation and genetic algorithms are implemented to resolve this behavior.

1.2 Speedup Summary

When using MPI, we achieved a speedup of ~2.8 when running across 4 threads. This was implemented by parallelizing the fitness function which takes an artificial neural network configuration and validates it to 1,000 test images.

1.3 Image Gallery

Evolution of 22 members across 100 generations. The color shown for each instance corresponds to the success rate in which the artificial neural network was able to classify test data.

Speedup factors across four threads.

Retrieved from "http://www.csi702.net/Classifying_Handwritten_Digits_with_the_Artificial_Neural_Network_and_Genetic_Algorithm_in_a_Distributed_Environment"
Personal tools