Versatile reservoir computing for heterogeneous complex networks

Can a small, versatile machine be trained on the time series of a few elements in a large, heterogeneous complex system, and then substitute any element while preserving its collective dynamics over a finite time horizon? Leveraging the technique of reservoir computing (RC) in machine learning, the authors demonstrate that such a machine can indeed be constructed, and performs effectively in general complex networks of heterogeneous dynamics and structure. The all-in-one, plug-and-play nature of versatile RC offers substantial convenience for maintaining the functionality of real-world complex systems that suffer sporadic or temporary component failures.