Multimedialne seminarium z ekono- i socjofizyki

Modelling spreading processes and diffusion on networks is one of the most popular subjects among researchers dealing with complex systems. The reason for that may be a growing number of phenomenon, which can be described with such models. Epidemics, fake news or cascading failures in power grids, to name only a few. In general, using these models in empirical setting is difficult because the spreading or transmission probabilities are not known. One way to estimate them is to use actual cascades and reverse engineer their values by maximising their likelihood. Unfortunately, in reality we are often able to observe only a fraction of the network, which makes this task computationally inefficient. We propose a novel efficient algorithm as a solution to this problem. Our approach is based on dynamic message passing and it allows for scalable computations, suited for large real-world networks. We present application of our method to the Independent Cascade model, but it can easily be generalised to other models.