We present a brief introduction to a flexible, general network inference framework which models data as a network space, sampled to optimize network structure to a particular task. We introduce a formal problem statement related to influence maximization in networks, where the network structure is not given as input, but learned jointly with an influence maximization solution.