Consensus sequence plots and error thresholds: tools for visualising the structure of fitness landscapes

Abstract

This paper investigates the occurrence of error thresholds in genetic algorithms (GAs) running on a wide range of fitness landscape structures. The error threshold, a notion from molecular evolution, is a critical mutation rate beyond which the evolutionary dynamics of a population changes drastically. The paper also introduces Consensus sequence plots, an empirical tool for locating error thresholds on complex landscapes. This plots were borrowed and adapted from theoretical biology. Results suggest that error thresholds occur in GAs but only on landscapes of certain degree of ruggedness or complexity. Moreover, consensus sequence plots can be useful for predicting some features of a landscape such as ruggedness and "stepness". We argue that error thresholds and consensus sequence plots, may become useful tools for analyzing evolutionary algorithms and visualising the structure of fitness landscapes.