Surrogate-Assisted Optimization of the Opt-IA Artificial Immune System Algorithm

Artificial immune systems are bio-inspired algorithms capable of optimizing highly difficult objective functions, with multi-modal and deceptive landscapes. In order to do so, they require large evaluation budgets. Here, we describe how to reduce the number of function evaluations of the Opt-IA artificial immune system algorithm through the use of a surrogate model based on Gaussian Processes. Furthermore, we describe an exploration strategy, based on the Sobol sequence, to bootstrap the optimization process. We show that such methods increase the convergence speed of Opt-IA on several functions on the noiseless BBOB testbed. Our method outperforms Opt-IA on most tested functions for low-precision targets.