Information flow in Gaussian processes across timescales

Information flow in Gaussian processes across timescales

This paper studies how information flows between layers in a multilayer network model where each layer has a different timescale. Using Gaussian processes and Fokker-Planck operators, the authors derive analytical solutions showing that edges between fast and slow layers do not generate information, but feedback from slow to fast layers does, which can then propagate along paths in the network topology. They apply this framework to study instability and find critical slow layers drive maximal information increases. The work generalizes past discrete process results to continuous systems, furthering understanding of how multilayer structure shapes emergent function.