Operationalizing Microsoft Cognitive Toolkit and TensorFlow models with HDInsight Spark

Sign in to queue

Description

Operationalizing deep neural networks (DNNs) with Azure HDInsight offers major advantages such as dynamic scalability, minimal data transfer latency, and the potential to implement both quasi-realtime and batch processing. Obstacles to implementation include deep learning framework installation on HDInsight clusters, matching the image preprocessing used during training without recourse to framework-specific file I/O procedures, and appropriately partitioning images across workers to minimize model load latency. This session demonstrates ways to address these issues: sample PySpark scripts for operationalizing both CNTK and TensorFlow DNNs are provided, with key approaches highlighted during the talk.

Day:

4

Level:

0

Track:

CE

Session Type:

Breakout: 45 minute

Code:

BRK3316

Room:

Hyatt Plaza International H

Embed

The Discussion

Add Your 2 Cents