Difficulty of linkage learning in estimation of distribution algorithms
Proceedings of the 11th Annual conference on Genetic and evolutionary computation
| Hi-index | 0.00 |
This work proposes a linkage-learning niching method that improves the capability of estimation of distribution algorithms (EDAs) on reducing spurious linkages which increase problems difficulty. Concatenated parity function (CPF), a class of allelic pairwise independent problems, causes exponential scalability for hierarchical Bayesian optimization algorithm (hBOA), which is one of powerful EDAs. Empirical results show that restricted tournament replacement (RTR) that hBOA employs results in spurious linkages and increases difficulty on solving CPF. Our research consists of these goals: (1) proposing a mutual information matrix to approximate the implicit linkage-information during EDAs' execution, (2) reducing spurious linkages by utilizing new metric of similarity, and (3) maintaining diversity of population. The results show that hBOA with our proposed niching method reduces the spurious linkages and solves CPF in the polynomial time.