Speaker
Description
Complex systems consist of interacting units connected through intricate networks. Predicting sudden changes in their dynamics is essential to mitigate the consequences of large-scale disruptions. This task is inherently challenging, as it requires forecasting behavior in parameter regimes where no data are available.
We address this problem for networks with chaotic local dynamics by reconstructing both the individual dynamics and a statistical description of their interactions directly from data. We show that the network behavior admits a decomposition into an emergent deterministic
component and a fluctuation term. While such fluctuations are traditionally treated as noise and filtered out, we demonstrate that they are in fact essential for uncovering the underlying interaction structure such as community structures. This enables the early prediction of synchronization transitions in networks with community structure, even when the system operates far from the transition regime.