Swift for TensorFlow: A next-generation framework for differential programming

Brennan Saeta (Google)

2:35pm–3:15pm Thursday, September 12, 2019

Location: Expo Hall 3

Implementing AI

Secondary topics: Deep Learning tools

Average rating:

(5.00, 2 ratings)

Who is this presentation for?

ML practitioners, engineers, and researchers

Level

Advanced

Description

If you’ve ever wanted to take the derivative for a dynamic program or Monte Carlo tree search, ever wanted to write your own kernels, or are sometimes bewildered by the massive API surface and complexity of modern deep learning frameworks, look no further than Swift for TensorFlow. The Swift for TensorFlow project is a next-generation framework for machine learning and differential computation. It includes a source-to-source autodifferentiation system built into the Swift language itself, resulting in a far more flexible and powerful toolchain.

Brennan Saeta leads you through why you should use Swift, the benefits of the Swift for TensorFlow project, and how to use Swift for TensorFlow in your own projects and applications.

Prerequisite knowledge

Familiarity with deep learning concepts

What you'll learn

Learn the strengths and weaknesses of Swift for TensorFlow, when you should consider it, and how to leverage the toolchain in your own projects, applications, and research

Brennan Saeta

Google

Brennan Saeta is a software engineer on the Google Brain team leading the Swift for TensorFlow project. Previously, he was the TensorFlow tech lead for Cloud TPUs.