Resource allocation between initialization and optimization under computational expensive environment

Abstract

Initialization techniques are normally considered as “resource-free” and their computational complexities are seldom addressed. Since many techniques require objective function evaluations to generate initial solutions, this “resource-free” assumption is invalid under computational expensive environment. In this paper, we propose an Computational Resource Optimization Problem (CROP) between initialization and optimization under such environment. We provide a comparison metric among different initialization techniques. Four popular initialization techniques, namely, Pseudo Random Number Generator (PRNG), Opposition-based Learning (OBL), Quasi-Opposition-based Learning (QOBL) and Quadratic Interpolation (QI) are studied. Differential Evolution (DE) is used as the underlying optimization technique, while Chemical Reaction Optimization (CRO) is used to solve CROP. The CEC2014 computational expensive problem set is used as test cases. Our results show the importance of considering resource allocation between initialization and optimization in computational expensive environment.