Differentiable Programming for Piecewise Polynomial Functions

Minsu Cho, Ameya Joshi, Xian Yeow Lee, Aditya Balu, Adarsh Krishnamurthy, Baskar Ganapathysubramanian, Soumik Sarkar, Chinmay Hegde

Keywords: Differentiable Programming, Piecewise polynomial regression, Generative models, Segmentation

TL;DR: We propose a novel approach to calculate weak Jacobians for piecewise polynomial functions, thus enabling their use in general differentiable programs.

Abstract: 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 of 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.