Meta-Learning-Assisted Constraint Relaxation for Constrained Black-Box Optimization
This letter proposes MeCO, a meta-learning-assisted optimizer that learns an adaptive $\epsilon$-relaxation policy for constrained BBO, and demonstrates evidence that MeCO transfers across held-out benchmark functions, higher dimensions, and an application-domain setting.
Si-Jie Ma, Ze-Yuan Ma, Yue-Jiao Gong et al.
· IEEE Computational Intellige... · 1 citation