Computing Expected Losses in Perturbation Models using Multidimensional Parametric Min-cuts

We consider the problem of computing and differentiating expected losses under perturbation-based probabilistic models. This is a challenging computational problem that has traditionally been tackled using Monte Carlo-based methods. In this work, we show how a generalization of parametric min-cuts can be used to address the same problem, achieving high accuracy faster than a sampling-based baseline.