COBRA++: Enhanced COBRA Optimizer with Augmented Surrogate Pool and Reinforced Surrogate Selection

Real-world optimization problems present significant challenges to optimization algorithms, such as the expensive evaluation issue and complex constraint conditions. The optimization of these problems can be defined as the form in section A.1 in appendix. COBRA optimizer is a representative tool for such optimization problems, which introduces RBF (Radial Basis Function) surrogate to reduce online evaluations and bi-stage optimization process. Though promising, its selection standard is still manually decided by human expert, resulting in labor-intensive fine-tuning for novel tasks. In this paper, we propose a learning-based adaptive strategy (COBRA++) that enhances COBRA in two aspects: 1) An augmented surrogate pool; 2) A reinforcement learning-based online model selection policy. We have conducted multi-dimensional validation experiments and demonstrate that COBRA++ achieves performance improvement against vanilla COBRA and its adaptive variant.