Bayesian inference of epidemic models on networks

by Alejandro Lage Castellanos (Univ. del Havana, Cuba e Politecnico di Torino)

Tuesday 2 Jul 2013, 12:00 → 14:00 Europe/Rome

Aula Rasetti (Dip. di Fisica - Edificio G. Marconi)

Diffusion on networks is an ubiquitous problem appearing in distant contexts like sociology, marketing, neural networks, computer networks, or diseases outbreaks. In this research we focus in the also interesting problem of inferring the origin of an epidemic in a graph. We focus on the Susceptible/Infected/Recovered model of epidemic diffusion, and implement a Bayesian method for the inference. It corresponds to the nontrivial application of Bethe/BP equations for the SIR model, and outperforms other old and recent approaches, being not only more accurate in its predictions, but also more flexible and formally justified.