Sociomeeting: Temporal networks as time series, high vs low dimensional representations

In this talk I will give an overview of some recent works aiming to study temporal networks from the lens of time series and dynamical systems. In the first part of the talk, we show how to extend time series and dynamic concepts such as the autocorrelation function or Lyapunov exponents to the temporal network realm, by which to infer memory and sensitivity to initial conditions of synthetic or real temporal network trajectories. Besides validating the methods, we illustrate these in the context of training of artificial neural networks. In the second part of the talk we ask the opposite question: how we can build a suitable scalar embedding of a temporal network trajectory? We show that the intrinsic dynamics of the network can thereby be efficiently characterised in the embedding space using simple one-dimensional time series analysis metrics. Potential applications will be briefly discussed.