APPLYING ARTIFICIAL NEURAL NETWORK TO DETECT UNIDIRECTIONAL COUPLING BETWEEN VAN DER POL OSCILLATORS FROM SHORT NOISY TIME SERIES

The report presents the result of applying a convolutional neural network to detect unidirectional coupling between two van der Pol oscillators in the absence of synchronization. The convolutional neural network was tested on short time series, up to 10 characteristic periods, in presence of measurement noise. The convolutional neural network, when compared to the correlation analysis, showed greater accuracy when detecting unidirectional coupling from short time series and better resistance to the measurement noise.