Saving computational budget in Bayesian network-based evolutionary algorithms

During the evolutionary process, algorithms based on probability distributions for generating new individuals suffer from computational burden due to the intensive computation of probability distribution estimations, particularly when using Probabilistic Graph Models (PGMs). In the Bayesian Optimisation Algorithm (BOA), for instance, determining the optimal Bayesian network structure by a given solution sample is an NP-hard problem. To overcome this issue, we consider a new BOA-based optimisation approach (FBOA) which explores the fact that patterns of PGM adjustments can be used as a guide to reduce the frequency of PGM updates because significant changes in PGM structure might not occur so frequently, and because they can be particularly sparse at the end of evolution. In the present paper, this new approach is scrutinised in the search space of an NK-landscape optimisation problem for medium and large-size instances. Average gaps and success rates as well as the correlation between the landscape ruggedness of the problem and the expected runtime of FBOA and BOA are presented for medium-size instances. For large-size instances, optimisation results from FBOA and BOA are compared. The experiments show that, despite our FBOA being of almost three times faster than BOA, it still produces competitive optimisation results.