https://doi.org/10.1140/epjs/s11734-025-01592-1
Regular Article
Application of neural networks to detection of unidirectional coupling between Van der Pol oscillators from ultrashort time series in the presence of noise
1
Saratov State University, Astrakhanskaya Str. 83, 410012, Saratov, Russia
2
Saratov Branch of the Institute of RadioEngineering and Electronics of Russian Academy of Sciences, Zelyonaya 38, 410019, Saratov, Russia
Received:
31
January
2025
Accepted:
14
March
2025
Published online:
31
March
2025
Detection of the directional coupling between complex systems is important both for practical applications and fundamental studies in various fields. However, traditional and common model-based methods, based on Granger causality or transfer entropy, require fitting autoregressive models to the time series of instantaneous phases. Precise description of complex systems and signals often requires high-order polynomials, which may be unstable. Computational complexity also makes model-based approaches less viable for real-time analysis, which is required in, for example, the rapidly developing field of personalized medicine. Therefore, it seems promising to adapt neural networks to the detection of coupling. We tested fully connected, convolution and recurrent neural networks on two Van der Pol oscillators with unidirectional coupling. We investigated resistance to measurement noises and minimal required duration of the time series. Our first results are positive and show that the neural networks can detect unidirectional coupling from very short time series and in the presence of intense noise. In our study, the fully connected neural network was more resistant to noise, but less sensitive to the weak coupling compared to the convolution network, which suggests that any particular neural network likely can’t provide a universal solution and must be tested on, for example, model data.
Copyright comment Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
© The Author(s), under exclusive licence to EDP Sciences, Springer-Verlag GmbH Germany, part of Springer Nature 2025
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
