|
Abstract:
|
Worldwide, the outbreak of the seasonal flu is responsible for more than three million cases of severe illness, over 250,000 deaths, and significant economic impacts every year. Management of the flu is difficult due to limited monitoring, lack of resources, and the massive population of susceptible individuals. Technological advancements in wearable devices offer tremendous promise for low-cost, widespread monitoring of disease progression and for identifying where, when, and to whom treatments should be applied. We model the flu as spreading across a large social network in discrete time and formalize a management strategy as a map from the current state of the epidemic to a subset of individuals recommended for treatment. An optimal management strategy minimizes cumulative time of infection for all the individuals in the network. We propose an estimator of the optimal management strategy based on Monte Carlo Tree Search that can scale to populations with millions of individuals. We also derive bounds on the trade-off between computational expenditures and solution quality. Our approach is illustrated with an agent-based model built from historical flu data.
|
