Wednesday, March 27, 2019

ML: Swift for TensorFlow (vs Python)

Swift is a modern programming language developed by Apple for mobile iOS and Mac.
And not it is being used by Google for AI development with TensorFlow!

How this happened? And what about Python?

Swift is developed by Chris Lattner who also created LLVM compiler platform.
He worked in Apple on compilers and tools, and he and his team created Swift as a modern open source replacement for Objective-C. Then he briefly worked in Tesla on self-driving, just to end up in Google Brain team.

TensorFlow is a hardware and software platform optimized for Machine Learning.
Usually it is used from Python, and that is a challenge, since Python is an interpreted language,
hard to closely integrate with low-level tool like TensorFlow.
Swift on the other side has performance of C++ (since it is using same compiler toolset)
and syntax and semantics of a very modern language!

In this case Swift can even be dynamically integrated with Python, to call existing Python libraries!
It can be used in interactive mode, compiled, or even from Jupyter Notebooks.
Very interesting!

In addition to mobile and desktop, Swift is also being used for "server side" development
and this is supported by IBM.

Swift for TensorFlow  |  TensorFlow

Swift for TensorFlow - TFiwS (TensorFlow Dev Summit 2018) - YouTube

Swift for TensorFlow (TensorFlow @ O’Reilly AI Conference, San Francisco '18) - YouTube

Exploring TensorFlow 2.0 with Paige Bailey - TWiML Talk #242

Coral | Home

SparkFun Edge Development Board - Apollo3 Blue - DEV-15170 - SparkFun Electronics

Introducing Swift For TensorFlow – TensorFlow – Medium

TensorFlow and Swift – QuarkWorks, Inc. – Medium

Swift creator Chris Lattner joins Google Brain after Tesla Autopilot stint | TechCrunch