Representation and Hidden Bias II: Eliminating Defining Length Bias in Genetic Search via Shuffle Crossover
نویسندگان
چکیده
The traditional crossover operator used in genetic search exhibits a position-dependent bias called the dcfining-length bias. We show how this bias results in hidden biases that are difficult to anticipate and compensate for. We introduce a new crossover operator, shuffle crossover, that eliminates the position dependent bias of the traditional crossover operator by shuffling the representation prior to applying crossover. We also present experimental results that show that shuffle crossover outperforms traditional crossover on a suite of five function optimization problems.
منابع مشابه
Effectiveness of neighborhood crossover in multiobjective genetic algorithm
In this paper, the effectiveness of the neighborhood crossover of EMO algorithms is discussed through the numerical experiments. The neighborhood crossover chooses two parents which are close to each other in the objective space. All the individuals are sorted in order of proximity in the objective space, and then neighborhood shuffle is conducted, which randomly replaces individuals in the pop...
متن کاملCrossover Bias in Genetic Programming
Path length, or search complexity, is an understudied property of trees in genetic programming. Unlike size and depth measures, path length directly measures the balancedness or skewedness of a tree. Here a close relative to path length, called visitation length, is studied. It is shown that a population undergoing standard crossover will introduce a crossover bias in the visitation length. Thi...
متن کاملSub-tree Swapping Crossover, Allele Diffusion and GP Convergence
We provide strong evidence that sub-tree swapping crossover when applied to tree-based representations will cause alleles (node labels) to diffuse within length classes. For a-ary trees we provide further confirmation that all programs are equally likely to be sampled within any length class when sub-tree swapping crossover is applied in the absence of selection and mutation. Therefore, we prop...
متن کاملVariable-Dimensional Optimization with Evolutionary Algorithms Using Fixed-Length Representations
This paper discusses a simple representation of variable-dimensional optimization problems for evolutionary algorithms. Although it was successfully applied to the optimization of multi-layer optical coatings, it is shown that it introduces a unintentional bias into the search process with respect to the probability of a dimension being generated by mutation and recombination. In order to exami...
متن کاملThe Distributional Biases of Crossover Operators
The choice of genetic operators is one way in which genetic algorithms can be tailored to speciic optimization problems. For bit represented problems, the choice of crossover operator , or the choice not to use a crossover operator, can dramatically aaect search performance. The eecacy of crossover for genetic search is governed by the relationship between the crossover biases and the search pr...
متن کامل