Computational & Technology Resources
an online resource for computational,
engineering & technology publications
Civil-Comp Proceedings
ISSN 1759-3433
CCP: 97
Edited by: Y. Tsompanakis, B.H.V. Topping
Paper 38

Introducing Full Memory in Genetic Algorithms

A.E. Charalampakis

National Technical University of Athens, Greece

Full Bibliographic Reference for this paper
A.E. Charalampakis, "Introducing Full Memory in Genetic Algorithms", in Y. Tsompanakis, B.H.V. Topping, (Editors), "Proceedings of the Second International Conference on Soft Computing Technology in Civil, Structural and Environmental Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 38, 2011. doi:10.4203/ccp.97.38
Keywords: genetic algorithms, external memory, evolutionary computation.

Evolutionary algorithms (EAs) are biologically inspired stochastic algorithms that have been successfully applied to a wide variety of problems [1]. The most widely known type of EAs are the genetic algorithms (GAs), which were initially conceived by Holland [2] and later employed in virtually any problem imaginable.

Most EAs utilize some kind of short- or long-term memory to explore the design space and produce better solutions. In the case of standard GA (SGA) the algorithm is practically memoryless, as it simply evolves a population of candidate solutions. Elitism is introduced as a short-term memory operator, linking the current to the previous generation only, which is shown to improve performance [3].

In this study, the concept of using full memory in GAs is examined. This is achieved by a fully encapsulated operator, called the "registrar", which is placed between the GA and the subroutine evaluating the objective function. The registrar records the results of all function evaluations and selectively replaces the individuals requested by the EA with similar ones taken from the registry, based on adaptive match criteria. Thus, function evaluations are avoided in an aggressive manner.

Preliminary results from the application of the registrar with the SGA, for a number of well known benchmark functions, clearly show that, first, the recording and exploitation of the results of all function evaluations is feasible for most real-life problems using modern-day computer systems; and second, significant increase in performance may be observed. This is evident even at the early stages of evolution, in accordance with the standard "birthday problem". Better improvement in performance was observed when defining the match criteria of the registrar in the genotypic level, in contrast to defining them in the phenotypic level. The encapsulation of the code also facilitates the implementation of the registrar with other EAs.

A.E. Eiben, J.E. Smith, "Introduction to evolutionary computing", Springer, New York, 2003.
J.H. Holland, "Adaptation in natural and artificial systems", University of Michigan Press, Ann Arbor, MI, 1975.
J.A. Vasconcelos, J.A. Ramírez, R.H.C. Takahashi, R.R. Saldanha, "Improvements in Genetic Algorithms", IEEE Transactions on Magnetics, 37(5), 3414-3417, 2001. doi:10.1109/20.952626

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £70 +P&P)