Differentiable Programming for Piecewise Polynomial Functions

Abstract

The paradigm of differentiable programming has considerably enhanced the scope of machine learning via the judicious use of gradient-based optimization. However, standard differentiable programming methods (such as autodiff) typically require that the models be differentiable, limiting their applicability. We introduce a new, principled approach to extend gradient-based optimization to piecewise smooth models, such as k-histograms, splines, and segmentation maps. We derive an accurate form to the weak Jacobian of such functions, and show that it exhibits a block-sparse structure that can be computed implicitly and efficiently. We show that using the redesigned Jacobian leads to improved performance in applications such as denoising with piecewise polynomial regression models, data-free generative model training, and image segmentation.

Tasks

Reproductions