Design, Implementation, Application, and Dissemination of an Evolving Expression-Linkage Genetic Algorithm
D. E. Goldberg*
U.S. Air Force Office of Scientific Research, F49620-97-1-0050

This work combines the best features of the fast messy genetic algorithm (fmGA), the gene expression messy genetic algorithm (gemGA), and the linkage learning genetic algorithm (LLGA) to create a broadly competent GA that solves an array of hard problems quickly, accurately, and reliably. Once developed, the so-called evolving expression-linkage GA (eelGA) will be applied to a problem of U.S. Air Force interest, and a series of seminars will be held to disseminate the methods and procedures to a variety of air force personnel and contractors.

Efficient Genetic Algorithms
D. E. Goldberg*
Illinois Genetic Algorithms Laboratory

This project seeks to obtain fast accurate solutions in genetic algorithms by: (1) spatial efficiencies, (2) temporal efficiencies, (3) sampling efficiencies and evaluation relaxations, and (4) systematic hybridization. The results of this project are important to the growing number of real-world applications of genetic and evolutionary computation.

Fast, Accurate Evolutionary Optimization Using an Extended Linkage Learning Genetic Algorithm
D. E. Goldberg*
IBM Corp., Shared University Research Program

An extended linkage learning genetic algorithm is developed and tested on hard combinatorial problems. The resulting algorithm should scale to solve boundedly difficult, misleading, interlocked, and poorly scaled problems in time that grows no more quickly than a polynomial function of the number of decision variables. These algorithms should be applicable to a range of practical problems in many disciplines of human endeavor.

The Modeling and Optimization of Parallel Genetic Algorithms
D. E. Goldberg,* E. Cantu-Paz
United States Information Agency-CONACYT, Fulbright Fellowship

This project builds on a recently developed model of population sizing for simple GAs to obtain models for bounding cases of parallel GAs. Our objective is to provide practitioners with simple ways to determine how to use their computational resources to find a solution of a desired quality. The bounding models cover the spectrum isolated subpopulations to completely connected subpopulations with maximal migration rates. They can be used to predict the parallel speedups attained with different communication schemes. The models also show that under some circumstances there is an optimal spatial allocation of individuals and that the expected execution time can be minimized.

Optimal Design Processor for Structural Systems
D. E. Goldberg*
National Science Foundation, INT 94-21297

This project funds travel and ancillary expenses to permit the principal investigator to visit the Indian Institute of Technology, Madras, and collaborate with C. S. Krishnamoorthy of the Department of Civil Engineering. The goal of the project is combine Illinois foundational and theoretical knowledge of genetic algorithms with IIT structural and GA applications experience to create a structural optimization system capable of solving difficult, large-scale structural problems.

Piecewise Development of Design Theory for Genetic Programming
D. E. Goldberg*
Illinois Genetic Algorithms Laboratory

Work at Illinois has resulted in the development of a piecewise theory of simple genetic algorithms that seeks to understand building block existence and definition, building block supply, building block difficulty, building block decision making, building block growth and timing, and building block mixing. This theory has been instrumental in (1) analyzing solution quality (or its lack thereof) in existing simple selectorecombinative GAs and (2) designing more effective crossover and ancillary operators. This project seeks to replicate the success of these efforts in the domain of genetic programming. Steps taken in the bit-string GE domain are retraced and enhanced as necessary for the understanding of the more complex situation of GP.