1 paper
Philipp Andelfinger, Justin N. Kreikemeyer
Recently proposed gradient estimators enable gradient descent over stochastic programs with discrete jumps in the response surface, which are not covered by automatic differentiati…