It's not really machine learning, it's just optimization (optimize the space of shapes to create the desired filter). The paper doesn't call it ML either