Boltzmann Weighted Selection Improves Performance of Genetic Algorithms

Abstract

We have implemented Boltzmann scaling on the optimization function to select the number of offspring each individual in the current population contributes to the next generation; the procedure outperforms a standard proportional scaling method on the small set of problems we have investigated. A broader range of problems should be used to test the generality of this result. The tolerance schedule is robust enough that the same schedule was used successfully for problems of different sizes and correspondingly different scales in optimization space. These results show that, for the molecular biology problem, many Boltzmann experiments completed with a correct solution before the decrease in tolerance that occurred after generation 10 and nearly all completed before the schedule leveled off again after generation 40.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Dec 01, 1991
Accession Number
ADA260155

Entities

People

  • Bruce Tidor
  • Michael De La Maza

Organizations

  • Massachusetts Institute of Technology

Tags

Communities of Interest

  • Energy and Power Technologies

DTIC Thesaurus Topics

  • Abstracts
  • Algorithms
  • Annealing
  • Artificial Intelligence
  • Convergence
  • Databases
  • Equations
  • Genetic Algorithms
  • Information Science
  • Information Systems
  • Machine Learning
  • Molecular Biology
  • Neural Networks
  • Optimization
  • Potential Energy
  • Probability
  • Standards

Readers

  • Adaptive Control and Estimation with Uncertainty in Dynamic Systems.
  • Applied Combinatorial Optimization and Logic Circuit Design.
  • Molecular and genetic basis of cancer.

Technology Areas

  • AI & ML
  • AI & ML - Machine Learning Algorithms
  • Biotechnology
  • Space