Analyzing the structure of cyclical competition using deep learning method

Abstract

We propose a new method using a deep neural network (DNN) to predict extinction in a cyclic competition model, which is motivated by the relationship between pattern formation and mobility. Numerical experiments showed that DNN performed well except for intermediate mobility, where the extinction probability increases sharply. Therefore, we suggest a hybrid method combining with Monte Carlo and DNN, whose computational cost reduces remarkably compared with the conventional method.