Abstract: A method for managing a communications network comprises receiving, as input to a multi-armed bandit, a decision instance defining a task for managing the communications network and environmental data. An arm of the multi-armed bandit is selected, wherein each arm comprises a model of a respective action from a plurality of possible actions that can be taken in response to the received decision instance. For the selected arm, execution of the associated action on the communications network is triggered to produce an outcome. The observed outcome of the selected arm is compared with an outcome predicted by the model of the selected arm. Using the comparison determine whether to update the model of the selected arm with a historic model, wherein the historic model defines a previously determined relationship between the action corresponding to the selected arm, the received decision instance and the environmental data.