SAFE: Scale-Adaptive Fitness Evaluation Method for Expensive Optimization Problems

Abstract

The key challenge of expensive optimization problems (EOP) is that evaluating the true fitness value of the solution is computationally expensive. A common method to deal with this issue is to seek for a less expensive surrogate model to replace the original expensive objective function. However, this method also brings in model approximation error. To efficiently solve the EOP, a novel scale-adaptive fitness evaluation (SAFE) method is proposed in this article to directly evaluate the true fitness value of the solution on the original objective function. To reduce the computational cost, the SAFE method uses a set of evaluation methods (EM) with different accuracy scales to cooperatively complete the fitness evaluation process. The basic idea is to adopt the low-accuracy scale EM to fast locate promising regions and utilize the high-accuracy scale EM to refine the solution accuracy. To this aim, two EM switch strategies are proposed in the SAFE method to adaptively control the multiple EMs according to different evolutionary stages and search requirements. Moreover, a neighbor best-based evaluation (NBE) strategy is also put forward to evaluate the solution according to its nearest high-quality evaluated solution, which can further reduce computational cost. Extensive experiments are carried out on the case study of crowdshipping scheduling problem in the smart city to verify the effectiveness and efficiency of the proposed SAFE method, and to investigate the effects of the two EM switch strategies and the NBE strategy. Experimental results show that the proposed SAFE method achieves better solution quality than some baseline and state-of-the-art algorithms, indicating an efficient method for solving EOP with a better balance between solution accuracy and computational cost.

Surrogate-Assisted Hybrid-Model Estimation of Distribution Algorithm for Mixed-Variable Hyperparameters Optimization in Convolutional Neural Networks

“…Furthermore, how to evolve the network depth and topology of CNNs is worthy of further research. Also, more efficient optimization paradigms can be considered for CNN optimization in different scenarios, where potential optimization paradigms include data-driven optimization [61], large-scale optimization [62], dynamic optimization [63], many-objective optimization [64], multi-modal optimization [65], expensive optimization [66], and parallel [67] and distributed optimization [68]. In addition, the SHEDA is potential for solving more challenging learning tasks in complex real-world applications, such as healthcare application [69] and autonomous robot application [70], which will be further explored and studied.…”

A Meta-Knowledge Transfer-Based Differential Evolution for Multitask Optimization

“…For future work, the proposed algorithm will be further extended to solving more difficult and complex MTOPs, such as not only in complex continuous space [54]- [56], but also in complex discrete [57]- [60], combinational [61]- [64], and mix-variable space [65]- [67]. Furthermore, as the MKT is a generic idea, further exploration of other kinds of meta-knowledge and other meta-knowledge transfer methods and utilization methods are worthy studied to obtain more powerful EMTO algorithms.…”

Evolutionary Computation for Expensive Optimization: A Survey

“…Lim et al [72] developed a dynamic fidelity computational model for FE, in which the fidelity of the computational model grows as the evolution progresses. Koziel [73] investigated a multi-fidelity optimization in which the computational model′s fidelity level can be adaptively modified. With simulation-based FE of different accuracy scales, Wu et al [74] developed a scale-adaptive FE (SAFE) approach for the crowdshipping scheduling application problem, which can strike a better balance between solution accuracy and computational cost.…”