Prof. Thomas M. Michelitsch
Sorbonne University, France

Title: Are compartment models implemented in AI helpful to improve predictability of real-world epidemics ?

Abstract: We propose a stochastic compartment model of epidemic spreading in complex networks with resetting and mortality for implementation in Artificial Intelligence (AI). We raise the question whether implementation of such a model in AI could help to improve predictability of real-world epidemics in order to open this discussion. We investigate an epidemic compartment model [1,2,3] where we account for mortality caused by the disease and stochastic relocations of individuals (random walkers) [4] We exclude demographic birth and death processes. Each individual is represented by a random walker which is in one of the states (compartments) S (susceptible for infection), E (exposed: infected but not infectious corresponding to the latency period), I (infected and infectious), R (recovered, immune), D (dead). We assume the time spans of sojourn in these compartments to be independent random variables drawn from specific distributions where we focus especially on Gamma distributions. In order to mimic human mobility patterns in real world structures such as cities we implement this model into a multiple random walker's approach. Each random walker performs an independent simple (Markovian) random walk on a connected complex random graph. A walker can only die during the period of its infection (while being in compartment I). We consider here diseases with direct transmission from I to S walkers. We explore the effects of stochastic relocations (stochastic resetting { SR [4]) of walkers to randomly selected nodes, mimicking the effects of long-range journeys on the spreading of the disease. We validate our model in random walk simulations on some kinds of random graphs (such as Barabasi-Albert (BA), Erdos-Renyi (ER) and Watts-Strogatz (WS)). Animated simulation videos of the spreading in a Watts-Strogatz graph without SR with mortality can be viewed online here and for the same setting without mortality here. For further information consult supplementary materials. Our model has applications in interdisciplinary contexts, such as certain chemical reactions, the propagation of wood fires, finance and population dynamics.

BIO: Full CV