Validation Check of Neural Network using Monte Carlo Stochastic Optimization Method (MOST)

Abstract

Recently, artificial intelligence has been applied in various fields, and the core optimization methods are required as an indispensable tool in engineering design, and the selection of high-precision optimization methods has become important. However, traditional optimization methods present challenges such as guaranteed convergence, convergence speed, and tuning constants. Therefore, we developed the Monte Carlo Stochastic Optimization (MOST), which is an optimization method using the Monte Carlo method. MOST first divides the search region into two parts and integrates them by Monte Carlo method. The integrals are then compared and the side with the smaller integral value is selected. The selected area is then split and integrated again in two, selecting the smaller side of the integration. By repeating the selection and integration in this way, the optimum solution is obtained. In this paper, we apply and validate the proposed method to optimize the weighting factors of neural networks for wine evaluation problem, used car identification problem, and abalone age determination problem.