Finding critical backbone structures with genetic algorithms

This paper introduces the concept of a critical backbone as a minimal set of variables or part of the solution necessary to be within the basin of attraction of the global optimum. The concept is illustrated with a new class of test problems Backbone in which the critical backbone structure is completely transparent. The performance of a number of standard heuristic search methods is measure for this problem. It is shown that a hybrid genetic algorithm that incorporates a descent algorithm solves this problem extremely efficiently. Although no rigorous analysis is given the problem is sufficiently transparent that this result is easy to understand. The paper concludes with a discussion of how the emergence of a critical backbone may be the salient feature in a phase transition from typically easy to typically hard problems. Categories and Subject Descriptors F.2 [Theory of Computation]: Analysis of Algoruthms and Problem Complexity. General Terms Algorithms, Theory Keywords Combinato...

• Type: Conference
• Year: 2007
• Where: GECCO
• Authors: Adam Prügel-Bennett
• Keywords: Critical Backbone, Critical Backbone Structure, GECCO 2007, Test Problems Backbone