An ODE-based neural network with Bayesian optimization

Abstract

An application of the Bayesian optimization to an ordinary differential equation-based neural network is proposed. The loss function was considered as a black box function of the coefficients, and Bayesian optimization was applied to obtain desirable parameter values. The proposed method drastically simplifies the implementation because the adjoint method-based updating of coefficients is not required. Numerical experiments demonstrate that the performance of the proposed method is comparable to that of existing methods.