Loss functions in the prediction of extreme events and chaotic dynamics using machine learning approach

In this paper we investigate the problem of prediction of chaotic dynamics and extreme events using an ensemble of deep neural networks. We test our framework on the artificial data generated using two systems containing high-amplitude outliers and extreme events: the Lienard system and two bursting Hindmarsh-Rose neurons with mutual chemical couplings. We study how the quality of the prediction depends on the type of loss function.