A study on Neural Network Optimization : The Comparison of Genetic Algorithm and Linear Programming = 뉴럴네트워크의 최적화 기법에 관한 연구 : 유전자 알고리즘과 선형계획법의 비교

국문 초록 (Abstract)

뉴럴 네트워크(Neural Network)기법은, 인간의 두뇌를 구성하고 있는 뇌세포의 상호작용과정을 모방하여, 복잡한 자연 및 사회현상과 관련된 문제를 해결하고자 개발된 컴퓨터 시스템이다. 이 기법 중 가장 많이 사용되는 방법으로 백프로퍼게이션(backpropagation)은 최적해를 얻는데, 해의 객관성 및 일관성에서 상당한 문제접이 있다고 지적받고 있다.
따라서 본 연구는 이런 단점을 보완하고자 개발된 두 가지의 최적화기법-유전자알고리즘(Genetic Algorithm)과 선형계획법(Linear Programming)에 기초한 뉴럴 네트워크기법을 문헌연구를 통해서 설명비교한다.

다국어 초록 (Multilingual Abstract)

The backpropagation is currently the most widely used optimization technique for training neural networks. However, it has been shown that this gradient technique is severely limited in its ability for finding predictable and consistent solutions. Since the solutions for backpropagation trained networks are highly dependent on user-defined parameters and there no known methods by which researchers can determine optimal configurations of these parameters, researchers have had to result to trial-and-error techniques in order to find appropriate parameter settings for good solutions. This type of research is very inefficient and is based primarily on chance.
Two recently proposed methods of neural network optimization, the genetic algorithm and linear programing, are less dependent on user-defined parameter settings and have been found to achieve respectable results. This result directly compares the genetic algorithm and linear programing for three classification problems that were recently introduced in linear programming literature.

목차 (Table of Contents)

• 1. INTRODUCTION
• 2. ARTIFICIAL NEURAL NETWORK
• 3. THE GENETIC ALGORITHM
• 4. LINEAR PROGRAMMING
• 5. ALGORITHM COMPARISON
• 6. CONCLUSIONS
• ·REFERENCES