Bolte, Jérôme and Pauwels, Edouard (2019) Conservative set valued fields, automatic differentiation, stochastic gradient methods and deep learning. Mathematical Programming. pp. 1-33. (In Press)

Preview

Text Download (486kB) | Preview

Abstract

Modern problems in AI or in numerical analysis require nonsmooth approaches with a
exible calculus. We introduce generalized derivatives called conservative
fields for which we develop a calculus and provide representation formulas. Functions having a conservative field are called path differentiable: convex, concave,
Clarke regular and any semialgebraic Lipschitz continuous functions are path differentiable. Using Whitney stratification techniques for semialgebraic and definable
sets, our model provides variational formulas for nonsmooth automatic diffrentiation oracles, as for instance the famous backpropagation algorithm in deep learning.
Our differential model is applied to establish the convergence in values of nonsmooth stochastic gradient methods as they are implemented in practice.