gLINC: identifying composability using group perturbation

We present two novel perturbation-based linkage learning algorithms that extend LINC [5]; a version of LINC optimised for decomposition tasks (oLINC) and a hierarchical version of oLINC (gLINC). We show how gLINC decomposes a fitness landscape significantly faster than both LINC and oLINC. We present details of LINC, oLINC and gLINC, an empirical comparison of their speed, accuracy and sensitivity to population size on a concatenated trap function, and a discussion of their complexity and correctness. Categories and Subject Descriptors F.2 [Analysis of algorithms and problem complexity]: General General Terms Algorithms, Experimentation Keywords Genetic Algorithms, Linkage Learning, Epistasis, Composition, Perturbation, Hierarchical

• Type: Conference
• Year: 2006
• Where: GECCO
• Authors: David Jonathan Coffin, Christopher D. Clack
• Keywords: GECCO 2006, Keywords Genetic Algorithms, LINC, Linkage Learning Algorithms, Optimization