David Malkin
ABSTRACT
A system composed of multiple interacting components is capable of responding to contextual information and producing a higher range of non-linear responses to stimuli compared to modular systems with a low degree of component interaction. However, the fitness landscape of highly integrated systems is more rugged indicating that such systems are likely to be less evolvable. In this work we use an artificial life simulation to investigate whether the evolvability of highly integrated systems can be improved if thelevel of integration between the system's components is underevolutionary control. When evolving our multi-component system we discover that the level of integration very quickly falls to virtually zero reducing the ruggedness of the landscape and making it nearly neutral. This allows the evolving population to explore the genome space without getting stuck on local optima. The components then integrate and the evolving population settles on the global optimum. This work is unique because the presented problem requires the evolving system to be fully integrated in order to solve it and as such the decreased ruggedness and near neutrality are not a permanent feature of the landscape but rather a property which is temporarily manipulated and exploited by the evolving population.
- Altenberg, L. Genome growth and the evolution of the genotype-phenotype map. Evolution and Biocomputation: Computational Models of Evolution, 889: 205--259. Google Scholar
Digital Library
- Bentley, P. J. & Corne, D. W. 2001. Creative evolutionary systems. Morgan Kaufmann Publishers Inc. San Francisco, CA, USA. Google ScholarDigital Library
- Bongard, J. C. & Paul, C. Making evolution an offer it can't refuse: Morphology and the extradimensional bypass. Proceedings of the 6th European Conference on Advances in Artificial Life, 401--412. Google ScholarDigital Library
- Cariani, P. A. Extradimensional bypass. Biosystems, 64(1-3): 47--53.Google ScholarCross Ref
- Ebner, M., Langguth, P., Albert, J., Shackleton, M., & Shipman, R. On neutral networks and evolvability. Evolutionary Computation, 2001.Proceedings of the 2001 Congress on, 1.Google Scholar
- Hordijk, W. A Measure of Landscapes. Evolutionary Computation, 4(4):335.Google Scholar
- Huynen, M. A. Exploring phenotype space through neutral evolution. Journal of Molecular Evolution, 43(3): 165--169.Google ScholarCross Ref
- Izquierdo-Torres, E. The Role of Nearly Neutral Mutations in the Evolution of Dynamical Neural Networks. Ninth International Conference of the Simulation and Synthesis of Living Systems (Alife 9), 322--327.Google Scholar
- Jones, T. & Forrest, S. Fitness Distance Correlation as a Measure of Problem Difficulty for Genetic Algorithms. Proceedings of the 6th International Conference on Genetic Algorithms Table of Contents, 184--192. Google ScholarDigital Library
- Kallel, L., Naudts, B., & Reeves, C. R. Properties of fitness functions and search landscapes. Natural Computing Series, 175--206.Google Scholar
- Mitchell, M. 1996. An introduction to genetic algorithms. MIT Press Cambridge, MA, USA. Google ScholarDigital Library
- Newman, M. E. J. Effects of selective neutrality on the evolution of molecular species. Proceedings: Biological Sciences, 265(1403): 1333--1338.Google Scholar
- Ohta, T. Evolution by nearly-neutral mutations. Genetica, 102: 83--90.Google Scholar
- Peretz, I. Modularity of music processing. Nature Neuroscience, 6(7): 688.Google Scholar
- Reidys, C. M. Combinatorial landscapes. SIAM Review, 44(1): 3. Google ScholarDigital Library
- Schuster, P., Fontana, W., Stadler, P. F., & Hofacker, I. L. From Sequences to Shapes and Back: A Case Study in RNA Secondary Structures. Proceedings: Biological Sciences, 255(1344): 279--284.Google Scholar
- Skellett, B., Cairns, B., Geard, N., Tonkes, B., & Wiles, J. Maximally rugged NK landscapes contain the highest peaks. Proceedings of the 2005 Conference on Genetic and Evolutionary Computation, 579--584. Google ScholarDigital Library
- Smith, T. Fitness Landscapes and Evolvability. Evolutionary Computation, 10(1): 1. Google ScholarDigital Library
- Smith, T., Philippides, A., Husbands, P., & O'Shea, M. Neutrality and ruggedness in robot landscapes.Google Scholar
- Stadler, P. F. Landscapes and their correlation functions. Journal of Mathematical Chemistry, 20(1): 1.Google ScholarCross Ref
- Toussaint, M. & Igel, C. Neutrality: A Necessity for Self-Adaptation. Arxiv Preprint Nlin.AO/0204038.Google Scholar
- Wagner, G. P. Complex adaptations and the evolution of evolvability. Evolution, 50(3): 967.Google Scholar
Index Terms
- Evolutionary benefits of evolvable component integration
Recommendations
A tunable model for multi-objective, epistatic, rugged, and neutral fitness landscapes
GECCO '08: Proceedings of the 10th annual conference on Genetic and evolutionary computationThe fitness landscape of a problem is the relation between the solution candidates and their reproduction probability. In order to understand optimization problems, it is essential to also understand the features of fitness landscapes and their ...
Selecting for evolvable representations
GECCO '06: Proceedings of the 8th annual conference on Genetic and evolutionary computationEvolutionary algorithms tend to produce solutions that are not evolvable: Although current fitness may be high, further search is impeded as the effects of mutation and crossover become increasingly detrimental. In nature, in addition to having high ...
Evolutionary dynamics on multiple scales: a quantitative analysis of the interplay between genotype, phenotype, and fitness in linear genetic programming
Redundancy is a ubiquitous feature of genetic programming (GP), with many-to-one mappings commonly observed between genotype and phenotype, and between phenotype and fitness. If a representation is redundant, then neutral mutations are possible. A ...
Comments