Brought to you by NumFOCUS Foundation and O’Reilly Media Inc.
The official Jupyter Conference
August 22-23, 2017: Training
August 23-25, 2017: Tutorials & Conference
New York, NY

Deep learning and Elastic GPUs using Jupyter

Tim Gasper (Bitfusion), Subbu Rama (Bitfusion)
1:50pm–2:30pm Thursday, August 24, 2017
Kernels
Location: Murray Hill Level: Intermediate
Average rating: **...
(2.00, 1 rating)

Who is this presentation for?

  • Data scientists, researchers, and developers leveraging Jupyter as part of their deep learning toolkits

Prerequisite knowledge

  • A gerneral understanding of deep learning concepts and using Jupyter for data science

What you'll learn

  • Understand how to leverage CPUs and GPUs with Jupyter for deep learning
  • Explore an modified kernel for easily using GPUs attached elastically over the network and quickly context switching between CPUs and GPUs

Description

Jupyter is an excellent interface for executing deep learning development and training. In fact, many of the tutorials that help you get started with deep learning frameworks use Jupyter notebooks because of the ability to provide small blocks of commented, executable code with easy reproducibility, ability to display intermediate feature extraction information, and useful metrics and console output.

Combined with GPUs, Jupyter makes for fast development and fast execution, but it is not always easy to switch from a CPU execution context to GPUs and back. Drawing on their work at Bitfusion, Tim Gasper and Subbu Rama share best practices for doing deep learning with Jupyter and explain how to work with CPUs and GPUs more easily by using network-attached Elastic GPUs and quick-switching between custom kernels.

Topics include:

  • Best practices for using Jupyter for deep learning
  • Examples of Jupyter deep learning using CPUs versus GPUs, including pros and cons and expected speed gains with GPUs
  • How to leverage Elastic GPUs in combination with CPU nodes
  • How a custom kernel combined with Elastic GPUs can make switching between CPUs and GPUs dead simple
Photo of Tim Gasper

Tim Gasper

Bitfusion

Tim Gasper is director of product and marketing at Bitfusion, a deep learning automation software company enabling easier, faster development of AI applications, and cofounder of Ponos, an IoT-enabled hydroponics farming technology company. Tim has over eight years of big data, IoT, and enterprise content product management and product marketing experience. He is a writer and speaker on entrepreneurship, the Lean Startup methodology, and big data analytics. Previously, Tim was global portfolio manager for CSC Big Data and Analytics, where he was responsible for the overall strategy, roadmap, partnerships, and technology mix for the big data and analytics product portfolio; VP of product at Infochimps (acquired by CSC), where he led product development for its market-leading open data marketplace and big data platform as a service; and cofounder of Keepstream, a social media analytics and curation company.

Photo of Subbu Rama

Subbu Rama

Bitfusion

Subbu Rama is cofounder and CEO at Bitfusion, a company providing tools to make deep learning and AI application development and infrastructure management faster and easier. Previously, Subbu held various roles at Intel, leading engineering efforts spanning design, automation, validation, and postsilicon. He worked on Intel’s first integrated-graphics CPU, Intel’s first low-power CPU, Atom, and its SOC, high-performance microservers (Intel’s first initiative on microservers using low-power mobile phone processors), and Xeon servers. Subbu also led Dell Innovation Labs, driving innovation and skunk works, later overseeing numerous new strategic technology initiatives at the intersection of software and the cloud. There he built an engineering team from the ground up and launched Dell’s first cloud infrastructure marketplace.