Iterative Domain Optimization (2005.10005v1)

Abstract: In this paper we study a new approach in optimization that aims to search a large domain D where a given function takes large, small or specific values via an iterative optimization algorithm based on the gradient. We show that the objective function used is not directly optimizable, however, we use a trick to approximate this objective by another one at each iteration to optimize it. Then we explore a use case of this algorithm in machine learning to find domains where the models output large and small values with respect of some constraints. Experiments demonstrate the efficiency of this algorithm on five cases with models trained on the titanic dataset.