Networked Restless Bandits with Positive Externalities

Restless multi-armed bandits are often used to model budget-constrained resource allocation tasks where receipt of the is associated with an increased probability a favorable state transition. Prior work assumes that individual arms only benefit if they receive directly. However, many occur within communities and can be characterized by positive externalities allow derive partial when their neighbor(s) resource. We thus introduce networked restless bandits, novel bandit setting in which both embedded directed graph. then present Greta, graph-aware, Whittle index-based heuristic algorithm efficiently construct constrained reward-maximizing action vector at each timestep. Our empirical results demonstrate Greta outperforms comparison policies across range hyperparameter values graph topologies. Code appendices available https://github.com/crherlihy/networked_restless_bandits.