Spatio-temporal causality for data-specific information modelling

Murilo Baptista (University of Aberdeen)

Causality lies at the heart of the scientific method. I will show in this talk that causality specifies how information about one variable past or future state in a system depends not only on the time-interval considered but also on the resolution of measurements taken on another variable. This spatio-temporal character of causality offers more data-specific ways of calculating information flow among variables from data, further allowing to make the best possible predictions about the states of variables in a system. This talk will then show how to exploit this dual nature of causality to create informational-theoretical modelings of real-world complex systems considering different configurations of accessible time-series or data sets. As examples of this approach, we will comment on recent results for the inference of the underlying network in neural networks or political-social-economic data.