Skip to Main content Skip to Navigation
Conference papers

A mathematical model for automatic differentiation in machine learning

Abstract : Automatic differentiation, as implemented today, does not have a simple mathematical model adapted to the needs of modern machine learning. In this work we articulate the relationships between differentiation of programs as implemented in practice and differentiation of nonsmooth functions. To this end we provide a simple class of functions, a nonsmooth calculus, and show how they apply to stochastic approximation methods. We also evidence the issue of artificial critical points created by algorithmic differentiation and show how usual methods avoid these points with probability one.
Complete list of metadata
Contributor : Edouard Pauwels <>
Submitted on : Tuesday, June 2, 2020 - 1:59:14 PM
Last modification on : Monday, December 14, 2020 - 6:08:14 PM
Long-term archiving on: : Wednesday, December 2, 2020 - 1:47:34 PM


Files produced by the author(s)


  • HAL Id : hal-02734446, version 1
  • ARXIV : 2006.02080


Jerome Bolte, Edouard Pauwels. A mathematical model for automatic differentiation in machine learning. Conference on Neural Information Processing Systems, Dec 2020, Vancouver, Canada. ⟨hal-02734446v1⟩



Record views


Files downloads