Fast and lean data science with TPUs

Martin Gorner (Google)

2:30pm–3:10pm Wednesday, October 30, 2019

Location: Grand Ballroom H

Accelerators

Average rating:

(4.00, 1 rating)

Who is this presentation for?

Developers and data scientists

Level

New to TensorFlow

Description

Google’s TPUs revolutionize the way data scientists work. Weeklong training times are a thing of the past, and many models can be trained in minutes in a notebook. Agility and fast iterations bring neural networks into regular software development cycles and many developers ramp up on machine learning.

Martin Gorner explores TPUs, then dives deep into the microarchitecture secrets behind them. You’ll learn how to use them in your day-to-day projects to iterate faster. In fact, Martin not only demos the model—he trains the models on stage in real time on TPUs.

Prerequisite knowledge

Familiarity with TensorFlow and neural networks

What you'll learn

Learn what TPUs are for and how to use them, the library of reference TPU-optimized models, and RetinaNet on TPU for object detection

Martin Gorner

Google

Martin Gorner is a developer advocate at Google, where he focuses on parallel processing and machine learning. Martin is passionate about science, technology, coding, algorithms, and everything in between. He spent his first engineering years in the Computer Architecture Group of ST Microelectronics, then spent the next 11 years shaping the nascent ebook market at Mobipocket, which later became the software part of the Amazon Kindle and its mobile variants. He’s the author of the successful TensorFlow Without a PhD series. He graduated from Mines Paris Tech.