ABSTRACT
Crossover is an important genetic operator that can combine beneficial genes together. Unfortunately, neuro-evolution (NE) has not experienced the benefits of crossover, despite significant efforts that enabled crossover for neural networks. Orthogonally, speciation has become an important feature in NE for diversity maintenance; however, speciation research has focused on what measure is driving speciation versus how the measure determines species. This research posits that an appropriate speciation heuristic can enable effective crossover in NE by determining potential mating partners. This paper investigates these concepts and presents empirical evidence that demonstrates; (1) the impact of the speciation heuristic, (2) crossover's negative effect, and (3) a speciation heuristic that enables effective crossover in NE.
- E. Mayr. The growth of biological thought: diversity, evolution, and inheritance. Harvard Univ. Press, 1982.Google Scholar
- J.-B. Mouret and S. Doncieux. Encouraging behavioral diversity in evolutionary robotics: An empirical study. Evolutionary computation, 20(1):91--133, 2012. Google ScholarDigital Library
- K. O. Stanley and R. Miikkulainen. Evolving neural networks through augmenting topologies. Evolutionary Computation, 10:99--127, 2002. Google ScholarDigital Library
- X. Yao. Evolving artificial neural networks. Proceedings of the IEEE, 87(9):1423--1447, 1999.Google ScholarCross Ref
Index Terms
- Impact of Speciation Heuristic on Crossover and Search in NEAT
Recommendations
Differential evolution using a superior---inferior crossover scheme
Differential evolution (DE) is a new population-based stochastic optimization, which has difficulties in solving large-scale and multimodal optimization problems. The reason is that the population diversity decreases rapidly, which leads to the failure ...
Effects of Speciation on Evolution of Neural Networks in Highly Dynamic Environments
LION 6: Revised Selected Papers of the 6th International Conference on Learning and Intelligent Optimization - Volume 7219Using genetic algorithms for solving dynamic optimization problems is an important area of current research. In this work, we investigate effects of speciation in NeuroEvolution of Augmenting Topologies NEAT, a well-known method for evolving neural ...
Multipopulation-Based Differential Evolution with Speciation-Based Response to Dynamic Environments
SEMCCO 2013: 4th International Conference on Swarm, Evolutionary, and Memetic Computing - Volume 8297Unlike static optimization problems, the position, height and width of the peaks may vary with time instances in dynamic optimization problems DOPs. Many real world problems are dynamic in nature. Evolutionary Algorithms EAs have been considered to ...
Comments