Application of multi-task learning in predicting synchronization

There are numerous indicators used to characterize the degree of synchronization for a non-identical system consisting of heterogeneous phase oscillators, such as the critical coupling of phase synchronization and the critical coupling of frequency synchronization and order parameter. Is it possible to predict these indicators simultaneously given the realistic situations of unknown system dynamics, including network structure, local dynamics, and coupling functions? This process, known as multi-task learning, can be achieved through the model-free technique of a feed-forward neural network in machine learning. To elaborate, we can measure the synchronization indicators of a limited number of allocation schemes and utilize these data to train the machine model. Once trained, the model can be employed to predict these indicators simultaneously for any novel allocation scheme. More importantly, the trained machine can also identify the optimal allocation for synchronization from a large pool of candidates. This method solves an outstanding question, which is how to allocate a given set of heterogeneous oscillators on a complex network in order to improve the synchronization performance. Leveraging multi-task learning's ability to predict multiple synchronization indicators, we can ensure that the system with the optimal performs well throughout the entire synchronization transition. Additionally, we test the scalability of the machine; one approach is to predict the indicators for a system composed of a new set of oscillators, and the other is to simultaneously predict the indicators of different systems.