We consider a problem of inferring contact network from nodal states observed during an epidemiological process. In a black-box Bayesian optimisation framework this problem reduces to a discrete likelihood optimisation over the set of possible networks. The high dimensionality of this set, which grows quadratically with the number of network nodes, makes this optimisation computationally challenging. Moreover, the computation of the likelihood of a network requires estimating probabilities of the observed data to realise during the evolution of the epidemiological process on this network. A stochastic simulation algorithm struggles to estimate rare events of observing the data (corresponding to the ground truth network) during the evolution with a significantly different network, and hence prevents optimisation of the likelihood. We replace the stochastic simulation with solving the chemical master equation for the probabilities of all network states. This equation also suffers from the curse of dimensionality due to the exponentially large number of network states. We overcome this by approximating the probability of states in the tensor-train decomposition. This enables fast and accurate computation of small probabilities and likelihoods. Numerical simulations demonstrate efficient black-box Bayesian inference of the network.
翻译:暂无翻译