Structural Causal Bandits: Where to Intervene?

We study the problem of identifying the best action in a sequential decisionmaking setting when the reward distributions of the arms exhibit a non-trivial dependence structure, which is governed by the underlying causal model of the domain where the agent is deployed.