Time-dependent collaboration schemes for cooperative coevolutionary algorithms

Cooperative coevolutionary algorithms are a popular approach to learning via problem decomposition. One important aspect of cooperative coevolutionary algo-rithms concerns how to select collaborators for com-puting the fitness of individuals in different populations. We argue that using a fixed number of collaborators dur-ing the entire search may be suboptimal. We experi-ment with a simple ad-hoc scheme that varies the num-bers of collaborators over time. Empirical comparisons in a series of problem domains indicate that decreasing the numbers of collaborators over time fares better than keeping the number fixed. We conclude with a brief dis-cussion of our findings and suggest directions for future research