Generalized Information Content based on Variability Map for Exploratory Landscape Analysis of Global Optimization Problems
T.A. Agasiev · Procedia Computer Science · 2019
Exploratory Landscape Analysis methods were developed to assess a variety of characteristic features of optimization problems, namely objective function features. These methods are widely used to distinguish and classify problems with respect to calculated vector of features in order to deeply explore optimization algorithm’s behavior. Results of the most landscape analysis methods depend on a sample of design points and corresponding objective values representing a discretized landscape of objective function. The Information Content of Fitness Sequences method being relatively stable to sampling methods variations additionally requires sample ordering. The paper introduces a new generalized approach to information content analysis based on aggregated variability map of function’s landscape. It gives more accurate and robust results without any ordering algorithm needed based on points generated by a chosen sampling method.