Software-based Automatic Differentiation is Flawed.

Daniel Johnson,Trevor Maxfield,Yongxu Jin,Ronald Fedkiw

Journal of Computational Physics（2025）SCI 2区SCI 1区

Stanford University

被引用7|浏览46

摘要

Various software efforts embrace the idea that object oriented programming enables a convenient implementation of the chain rule, facilitating so-called automatic differentiation via backpropagation. Such frameworks have no mechanism for simplifying the expressions (obtained via the chain rule) before evaluating them. As we illustrate below, the resulting errors tend to be unbounded.

更多

查看译文

关键词

Machine Learning,Automatic Differentiation