Differentiable programming: Difference between revisions

Revision as of 18:01, 24 June 2019

Revision as of 17:14, 27 June 2019

Line 13:

A more recent package in the [[Julia (programming language)|Julia]] programming language — [https://github.com/FluxML/Zygote.jl Zygote] — resolves the issues that earlier attempts faced by treating the language's syntax as the graph; the design of the Julia language makes it easy for the [[intermediate representation]] of arbitrary Julia code to be differentiated directly, [[compiler optimization|optimized]], and compiled.<ref name="flux" /><ref>{{cite arxiv|last=Innes|first=Michael|date=2018-10-18|title=Don't Unroll Adjoint: Differentiating SSA-Form Programs|eprint=1810.07951|class=cs.PL}}</ref> An in-development differentiable programming language called [[Myia (programming language)|Myia]] also uses a similar approach.<ref name="myia1" />

Differentiable programming in Swift with TensorFlow extends the type system to make differentiable functions first-class. values,It andleverages [[Protocol (object-oriented programming)|protocol-oriented programming]] ([[Type class|type classes]]) to enable custom differentiable data structures. Automatic differentiation is implemented as a compiler transformation on the Swift intermediate language ([https://github.com/apple/swift/blob/tensorflow/docs/SIL.rst SIL]). ItThe leveragesauthors [[Protocol (object-oriented programming)|protocol-oriented programming]] ([[Type class|type classes]])plan to allow custompitch differentiable dataprogramming structures.as Thean authors hope that it will become a fully integrated part of theofficial Swift language in the futurefeature.<ref>{{Cite web|url=https://forums.swift.org/t/pre-pre-pitch-swift-differentiable-programming-design-overview/25992|title=Pre-pre-pitch: Swift Differentiable Programming Design Overview|date=2019-06-17|website=Swift Forums|language=en-US|access-date=2019-06-18}}</ref>

== APIs==

Many programming languages whose design is less amenable to differentiable programming can gain differentiability features by calling on a differentiable programming framework via an [[Application programming interface|API]]. For example, [[TensorFlow]] can be called from [[Python (programming language)|Python]], [[JavaScript]], [[C++]], [[Java (programming language)|Java]], [[Go (programming language)|Go]], and [[Swift (programming language)|Swift]].<ref>{{Cite web|url=https://www.tensorflow.org/api_docs|title=API Documentation {{!}} TensorFlow Core r1.14|website=TensorFlow|language=en|access-date=2019-06-20}}</ref>

Differentiable programming in Swift with TensorFlow extends the type system to make differentiable functions first-class values, and is implemented as a compiler transformation on the Swift intermediate language ([https://github.com/apple/swift/blob/tensorflow/docs/SIL.rst SIL]). It leverages [[Protocol (object-oriented programming)|protocol-oriented programming]] ([[Type class|type classes]]) to allow custom differentiable data structures. The authors hope that it will become a fully integrated part of the Swift language in the future.<ref>{{Cite web|url=https://forums.swift.org/t/pre-pre-pitch-swift-differentiable-programming-design-overview/25992|title=Pre-pre-pitch: Swift Differentiable Programming Design Overview|date=2019-06-17|website=Swift Forums|language=en-US|access-date=2019-06-18}}</ref>

==See also==