Swift for TensorFlow (TF World '19)

Swift for TensorFlow is a next generation machine learning platform that leverages innovations like first-class differentiable programming to seamlessly integrate deep neural networks with traditional AI algorithms and general purpose software development. In this session, learn how Swift for TensorFlow can make advanced machine learning research easier and faster.

Presented by: Paige Bailey, Brennan Saeta

[koendejonghe]

Great presentation, thank you. 27:23 I think that should be 'return dense2(tmp + tmp2)'

[geoffreyanderson]

Is the new LLVM-owned back-end cross-compiler the reason for this new HLL? Google gave over something of a cross-compiler to the LLVM project recently for outputting instructions to variety of processors including TPU, GPU, CPU, FPGA(?), etc.

[alkhashtee]

Thank you very much for the great presentation, but how we can participate this kind of activity?

[geoffreyanderson]

Why don't you just change python to be built on Swift atoms instead of C atoms? You'd make it easier on the humans -- and the Jupyter Notebook project and the other development IDEs, and not to mention the existing investment Google already made in documentation and tutorials for tensorflow using the python host language which are never going to disappear from Google search results it seems -- and you'd still get your LLVM and the {CPU, GPU, TPU, FPGA, etc}-compiler that Google handed to LLVM to be happy too. Of course there's the problem of the mountains of useful python libraries that already exist, which present a problem for reengineering the python language. But then that's no more difficult than writing all these over again just for Swift. FUrther, you might be able to automate the python library transformation to the new compiler backend.

[geoffreyanderson]

Were C#, Objective-C, Julia, Rust, Go, Java, Kotlin, and Scala not able to do this job? Was yet another language needed?