Code Growth, Explicitly Defined Introns, and Alternative Selection Schemes

Smith, Pete; Harries, Kim

doi:10.1162/evco.1998.6.4.339

Cited by 37 publications

(27 citation statements)

References 2 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Compact solutions are thought to be more robust and generalize better [27,49,61,77,81]. Introns also do seem to provide some protection against the destructive effects of crossover and other genetic operators [1,11,49,70,72] although this may not always be helpful. The usage of explicitly defined artificial introns has yielded generally good results in linear GP [38,50,51], but in tree-based GP it usually degraded the performance of the search process [3,10,70].…”

Section: Bloatmentioning

confidence: 99%

Dynamic limits for bloat control in genetic programming and a review of past and current bloat theories

Silva

Costa

2009

Genet Program Evolvable Mach

137

View full text Add to dashboard Cite

Bloat is an excess of code growth without a corresponding improvement in fitness. This is a serious problem in Genetic Programming, often leading to the stagnation of the evolutionary process. Here we provide an extensive review of all the past and current theories regarding why bloat occurs. After more than 15 years of intense research, recent work is shedding new light on what may be the real reasons for the bloat phenomenon. We then introduce Dynamic Limits, our new approach to bloat control. It implements a dynamic limit that can be raised or lowered, depending on the best solution found so far, and can be applied either to the depth or size of the programs being evolved. Four problems were used as a benchmark to study the efficiency of Dynamic Limits. The quality of the results is highly dependent on the type of limit used: depth or size. The depth variants performed very well across the set of problems studied, achieving similar fitness to the baseline technique while using significantly smaller trees. Unlike many other methods available so far, Dynamic Limits does not require specific genetic operators, modifications in fitness evaluation or different selection schemes, nor does it add any parameters to the search process. Furthermore, its implementation is simple and its efficiency does not rely on the usage of a static upper limit. The results are discussed in the context of the newest bloat theory.

show abstract

Section: Bloatmentioning

confidence: 99%

Dynamic limits for bloat control in genetic programming and a review of past and current bloat theories

Silva

Costa

2009

Genet Program Evolvable Mach

137

View full text Add to dashboard Cite

show abstract

“…Smith and Harries have shown that growth can occur in code that does influence fitness if the code has only a negligible effect on performance [14]. Soule et al have shown that code growth can also occur in exons that have a significant effect on fitness [15] [19].…”

Section: Introductionmentioning

confidence: 99%

Comparing genetic robustness in generational vs. steady state evolutionary algorithms

Jones

Soule

2006

Proceedings of the 8th Annual Conference on Genetic and Evolutionary Computation

View full text Add to dashboard Cite

Previous research has shown that evolutionary systems not only try to develop solutions that satisfy a fitness requirement, but indirectly attempt to develop genetically robust solutions as wellsolutions where average loss of fitness due to crossover and other genetic variation operators is minimized. It has been shown that in a simple "two peaks" problem, where the fitness landscape consists of a broad, low peak, and a narrow, high peak, individuals initially converge on the lower (less fit), but broader peak, and that increasing an individual's genetic robustness through growth is a necessary prerequisite for convergence on the higher, narrower peak [18]. If growth is restricted, the population remains converged on the less fit solution. We tested whether this result holds true only for generational algorithms, or whether it applies to steady state algorithms as well. We conclude that although growth occurs with both algorithms, the steady state algorithm is able to converge on the higher peak without this growth. This result shows that the role of genetic robustness in the evolutionary process is significantly different in generational versus steady state algorithms.

show abstract

“…However, Luke [10] has argued that introns themselves are not the cause of code growth. Smith and Harries have shown that growth can occur in code that does influence fitness (exons) if the exons only have a negligible effect on performance [17]. More recently, Soule has shown that code growth can occur even with exons that have a significant impact on the programs' fitness [18].…”

Section: Introductionmentioning

confidence: 99%

“…Nordin, Francone, and Banzhaf introduced a taxonomy of intron types for their linear GP system with 5 code categories based on whether changing the given code could affect the program's behavior [14]. Smith and Harries adopted this taxonomy for tree structured GP [17] …”

Section: Introductionmentioning

confidence: 99%

Growth of self-canceling code in evolutionary systems

Zhang

Soule

2006

Proceedings of the 8th Annual Conference on Genetic and Evolutionary Computation

View full text Add to dashboard Cite

This research examines the behavior of inoperative code (introns) in the evolution of genetically robust solutions. Genetically robust solutions are solutions that are less likely to be degraded by genetic operators, such as crossover. Previous work has shown that there is significant evolutionary pressure in favor of genetically robust solutions and that evolving programs adopt a number of strategies to increase genetic robustness, notably an increase in inoperative 'genes' (individual genetic units that don't influence fitness) and a preference for 'genes' with a relatively small effect on fitness.Here we examine the role of genes that cancel each other out. We find that allowing such 'canceling genes' leads to an overall increase in the rate of code growth, both through the inclusion of self-canceling code and through a general increase in introns. Finally, we find that the evolution generally follows a two-step process. Initially the operative code evolves rapidly to achieve a (near) optimal fitness. Then, the inoperative code begins to evolve most rapidly to increase robustness. In an extreme case of a problem that can be solved with no operative genes, individuals evolve by losing all operative genes and then losing all inoperative genes.

show abstract

Code Growth, Explicitly Defined Introns, and Alternative Selection Schemes

Cited by 37 publications

References 2 publications

Dynamic limits for bloat control in genetic programming and a review of past and current bloat theories

Dynamic limits for bloat control in genetic programming and a review of past and current bloat theories

Comparing genetic robustness in generational vs. steady state evolutionary algorithms

Growth of self-canceling code in evolutionary systems

Contact Info

Product

Resources

About