A new optimal resource allocation scheme for computationally expensive problems

Abstract

Initialization will affect the performance of Evolutionary Algorithm (EA), especially under computationally expensive environments. To investigate the optimal resource allocation between the initialization and optimization stages, we studied the Computational Resource Optimization Problem (CROP) in our previous work. In this paper, we extend our resource allocation model by separating “initial solutions” into two sets, namely, Generated Initial Solutions (GIS) and Used Initial Solutions (UIS). Such separation in the model allows us to conduct a more comprehensive analysis and reveal more insights into how initial solutions can affect the performance of EAs under computationally expensive environments. Simulations are conducted on the BBOB'09 benchmark functions with the standard settings to mimic the expensive environment. We have shown that proper computational resource allocation can improve the solution quality. Our analysis allows us to find the best allocation scheme for a particular EA on a specific problem and also demonstrates the importance of initialization.