摘要
We propose a model for studying the statistical properties of adaptive walks on correlated fitness landscapes which are established in genotype spaces of complex structure. The fitness distribution on the genotype space follows either the bivariate Gaussian distribution or the bivariate exponential distribution. In both cases the degree of correlation of the fitness landscape can be tuned by using a single parameter. To perform the adaptive walks two distinct rules are applied: the random adaptation walk (RAW) and the gradient adaptation walk (GAW). While for the RAW the mean walk length, (L) over bar, is a monotonic increasing function of the connectivity of the genotype space, for the GAW (L) over bar is a one-humped function. The RAW produces longer adaptive walks compared to the GAW, though its performance is slightly poorer and thereby the local maxima reached by the GAW algorithm are usually closer to the global optimum of the fitness landscape.
- 出版日期2012-2