Prediction of extreme events and chaotic dynamics using WaveNet
Autors:
Nikolay Gromov , Lev Smirnov , Tatiana Levanova ,
Pages:
20-31
Annotation:
In this paper we propose an approach for chaotic time series and extreme events prediction based on WaveNet model, which is a deep neural network that directly synthesizes speech waveforms from acoustic features. We test our approach on artificial data obtained from long time series with extreme events generated by two coupled bursting Hindmarsh-Rose neurons, and on two real-life data sets (local field potentials recorded in mice and electroencephalogram recorded in humans) containing patterns of epileptiform activity, which also can be viewed as extreme events.
File (pdf):