NIPS 2018 Expo Demo

Dec. 2, 2018

Expo Schedule »

Accelerating TensorFlow performance with nGraph

Sponsor: Intel AI

Organizers:
Harry Kim (Intel AI)

Presenters:
Harry Kim (Intel AI), Avijit Chakraborty (Intel AI)

Abstract:

nGraph is an open source compiler, runtime and computational library for deep learning acceleration. It enables data scientists to develop AI algorithms with their favorite deep learning framework and deploy them on to commodity hardware with minimal code change. In addition to multiple framework and hardware support, nGraph accelerates deep learning computation with state of art computational graph & device specific optimizations.

In this demo, we plan to show how nGraph can accelerate training & inference of a simple vision workload on TensorFlow. Participants are encouraged to bring their own laptop running on either Linux & Mac OS. We will provide them step by step instructions to install and use nGraph package with TensorFlow. We demonstrate the performance boost from executing deep learning workload with nGraph compared to native TensorFlow implementation.

·