Using a neural network to determine fitness in genetic design

Shane M Farritor, Jun Zhang

Research output: Contribution to conferencePaper

2 Citations (Scopus)

Abstract

Many automated design approaches require an objective function to determine the quality of a given design. Often, this function depends on a complex relationship between many parameters. Some parameters may be subjective and the relationships difficult to quantify. This paper presents a method where a neural network is used to evaluate the quality of proposed designs during a genetic algorithm search. In general application of the approach, a human designer would propose candidate designs for a given problem. These candidate designs are used to train a neural network fitness function. Then the genetic algorithm evolves new designs that the human designer might not conceive. In this way, the proposed approach would aid in the brainstorming process. The method is applied to the genetic design of modular robots for planetary exploration. This application is briefly described and the genetic design method is summarized. Then the neural network structure is explained and the training method is detailed. Finally, the neural network is used with the genetic design method to create a robot for a specific task.

Original languageEnglish (US)
Pages417-423
Number of pages7
StatePublished - Dec 1 2001
Event2001 ASME Design Engineering Technical Conference and Computers and Information in Engineering Conference - Pittsburgh, PA, United States
Duration: Sep 9 2001Sep 12 2001

Conference

Conference2001 ASME Design Engineering Technical Conference and Computers and Information in Engineering Conference
CountryUnited States
CityPittsburgh, PA
Period9/9/019/12/01

Fingerprint

Fitness
Neural Networks
Neural networks
Design Method
Robot
Genetic Algorithm
Genetic algorithms
Modular robots
Fitness Function
Design
Network Structure
Quantify
Objective function
Evaluate
Robots

ASJC Scopus subject areas

  • Modeling and Simulation
  • Mechanical Engineering
  • Computer Science Applications
  • Computer Graphics and Computer-Aided Design

Cite this

Farritor, S. M., & Zhang, J. (2001). Using a neural network to determine fitness in genetic design. 417-423. Paper presented at 2001 ASME Design Engineering Technical Conference and Computers and Information in Engineering Conference, Pittsburgh, PA, United States.

Using a neural network to determine fitness in genetic design. / Farritor, Shane M; Zhang, Jun.

2001. 417-423 Paper presented at 2001 ASME Design Engineering Technical Conference and Computers and Information in Engineering Conference, Pittsburgh, PA, United States.

Research output: Contribution to conferencePaper

Farritor, SM & Zhang, J 2001, 'Using a neural network to determine fitness in genetic design' Paper presented at 2001 ASME Design Engineering Technical Conference and Computers and Information in Engineering Conference, Pittsburgh, PA, United States, 9/9/01 - 9/12/01, pp. 417-423.
Farritor SM, Zhang J. Using a neural network to determine fitness in genetic design. 2001. Paper presented at 2001 ASME Design Engineering Technical Conference and Computers and Information in Engineering Conference, Pittsburgh, PA, United States.
Farritor, Shane M ; Zhang, Jun. / Using a neural network to determine fitness in genetic design. Paper presented at 2001 ASME Design Engineering Technical Conference and Computers and Information in Engineering Conference, Pittsburgh, PA, United States.7 p.
@conference{a7b23dc6e8a841ef9bb43b3a0f55bb6b,
title = "Using a neural network to determine fitness in genetic design",
abstract = "Many automated design approaches require an objective function to determine the quality of a given design. Often, this function depends on a complex relationship between many parameters. Some parameters may be subjective and the relationships difficult to quantify. This paper presents a method where a neural network is used to evaluate the quality of proposed designs during a genetic algorithm search. In general application of the approach, a human designer would propose candidate designs for a given problem. These candidate designs are used to train a neural network fitness function. Then the genetic algorithm evolves new designs that the human designer might not conceive. In this way, the proposed approach would aid in the brainstorming process. The method is applied to the genetic design of modular robots for planetary exploration. This application is briefly described and the genetic design method is summarized. Then the neural network structure is explained and the training method is detailed. Finally, the neural network is used with the genetic design method to create a robot for a specific task.",
author = "Farritor, {Shane M} and Jun Zhang",
year = "2001",
month = "12",
day = "1",
language = "English (US)",
pages = "417--423",
note = "2001 ASME Design Engineering Technical Conference and Computers and Information in Engineering Conference ; Conference date: 09-09-2001 Through 12-09-2001",

}

TY - CONF

T1 - Using a neural network to determine fitness in genetic design

AU - Farritor, Shane M

AU - Zhang, Jun

PY - 2001/12/1

Y1 - 2001/12/1

N2 - Many automated design approaches require an objective function to determine the quality of a given design. Often, this function depends on a complex relationship between many parameters. Some parameters may be subjective and the relationships difficult to quantify. This paper presents a method where a neural network is used to evaluate the quality of proposed designs during a genetic algorithm search. In general application of the approach, a human designer would propose candidate designs for a given problem. These candidate designs are used to train a neural network fitness function. Then the genetic algorithm evolves new designs that the human designer might not conceive. In this way, the proposed approach would aid in the brainstorming process. The method is applied to the genetic design of modular robots for planetary exploration. This application is briefly described and the genetic design method is summarized. Then the neural network structure is explained and the training method is detailed. Finally, the neural network is used with the genetic design method to create a robot for a specific task.

AB - Many automated design approaches require an objective function to determine the quality of a given design. Often, this function depends on a complex relationship between many parameters. Some parameters may be subjective and the relationships difficult to quantify. This paper presents a method where a neural network is used to evaluate the quality of proposed designs during a genetic algorithm search. In general application of the approach, a human designer would propose candidate designs for a given problem. These candidate designs are used to train a neural network fitness function. Then the genetic algorithm evolves new designs that the human designer might not conceive. In this way, the proposed approach would aid in the brainstorming process. The method is applied to the genetic design of modular robots for planetary exploration. This application is briefly described and the genetic design method is summarized. Then the neural network structure is explained and the training method is detailed. Finally, the neural network is used with the genetic design method to create a robot for a specific task.

UR - http://www.scopus.com/inward/record.url?scp=1542751053&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=1542751053&partnerID=8YFLogxK

M3 - Paper

SP - 417

EP - 423

ER -