Machine learning applied to pattern characterization in spatially extended dynamical systems

This new tool, in the form of Machine Learning (ML), has proven to be very useful in several areas of physics, due to its strong versatility, its ability to obtain patterns in very complex systems. In this work we explore techniques from Machine Learning (ML) to characterize spatio-temporal patterns in complex dynamical systems. These techniques are applied in coupled map lattices, for which the relevant parameters are the nonlinearity and coupling strength. As a training phase of our ML, we show several samples with the dynamic characteristics of each known space-time profile, such as frozen random pattern, pattern selection, chaotic defects, intermittency and fully developed space-time chaos, for example. After the training phase, we apply our algorithm to different values of non-linearity and coupling, where given the dynamic characteristics, for each pair of parameters, we can accurately identify the regions where each of these profiles is formed. (C) 2021 Elsevier B.V. All rights reserved.