We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.PL

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Programming Languages

Title: A Simple Differentiable Programming Language

Abstract: Automatic differentiation plays a prominent role in scientific computing and in modern machine learning, often in the context of powerful programming systems. The relation of the various embodiments of automatic differentiation to the mathematical notion of derivative is not always entirely clear---discrepancies can arise, sometimes inadvertently. In order to study automatic differentiation in such programming contexts, we define a small but expressive programming language that includes a construct for reverse-mode differentiation. We give operational and denotational semantics for this language. The operational semantics employs popular implementation techniques, while the denotational semantics employs notions of differentiation familiar from real analysis. We establish that these semantics coincide.
Comments: In POPL2020
Subjects: Programming Languages (cs.PL); Machine Learning (cs.LG)
DOI: 10.1145/3371106
Cite as: arXiv:1911.04523 [cs.PL]
  (or arXiv:1911.04523v2 [cs.PL] for this version)

Submission history

From: Gordon Plotkin [view email]
[v1] Mon, 11 Nov 2019 19:14:15 GMT (98kb)
[v2] Mon, 18 Nov 2019 23:02:37 GMT (98kb)

Link back to: arXiv, form interface, contact.