Reformulation and sampling to solve a stochastic network interdiction problem