Application of Monte Carlo stochastic optimization (MOST) to deep learning

In this paper, we propose a new optimization method based on the Monte Carlo method. The proposed is applied to several benchmark problems, and result of applying it neural network reported. Deep machine learning using networks one important keywords promote innovation in today’s advanced information society. Therefore, research large-scale, high-speed, high-precision algorithms has been actively conducted. author developed an which search region for multivariate parameters constituting objective function divided into two regions each parameter, integral values are numerically calculated by method, magnitude value compared, optimum point judged be small region. was 50 variable functions (Schwefel Ackley Functions), compared with results genetic algorithm (GA) representative existing methods. As result, confirmed that faster more accurate than GA. addition, networks, specifically XOR gate circuits IRS classification verified. optimized MOST reproduced teacher data test accurately conventional Adam algorithms.