Tree structure based differential evolution for optimization of trees and interactive evolutionary computation

A new Differential Evolution (DE) technique called Tree Structure Based Differential Evolution (TSDE) is proposed for an Interactive Evolutionary Computation (IEC). In IEC, repetitious evaluations by the user is used for optimization, therefore user´s fatigue is a practical issue. IEC needs to discover a solution with smaller population size and fewer search generations. DE turns out to be suited to IEC because of less user´s fatigue by pairwise comparison and effective search using difference vectors of two individuals. However, DE was originally proposed as a technique for continuous-valued optimization. Tree structure optimization problems can be solved with these advantages of DE if DE is extended for tree structure optimization problems. In TSDE, difference between two trees is defined as the set of parts that appear only in one of the two trees, and based on this, appropriate scaling and mutation operations are devised. The authors show the effectiveness of TSDE as an IEC technique by compared with other conventional methods for tree structure optimization problems in simulations.