SLIDES: Converged and Containerized Distributed Deep Learning With TensorFlow and Kubernetes

File uploaded by slimbaltagi on Sep 30, 2017
Version 1Show Document
  • View in full screen mode

Slides of a talk by Mathieu Dumoulin at the NYC Advanced Analytics Meetup hosted at McKinsey and sponsored by Hexstream: Distributed Deep Learning with TensorFlow in Docker containers - New York City (NYC) Advanced Analytics Meetup (New Yor… 

MapR recently announced the MapR XD platform, a global data fabric. It is a logical continuation of Convergence, which merges into a single cluster all of the MapR platform's technology in distributed file system, NoSQL and Document DB as well as real time event streams (Kafka). The sum of these technologies is definitely greater than the parts.

In this talk, we'll look at convergence in action using distributed deep learning as an example. First, we're going to make use of the MapR Persistent Application Client Container (PACC) to demonstrate distributed Tensorflow running within docker containers on data stored on MapR. 

Then, using the Deep Learning example as a base, we'll show how the unique features of MapR can be cleverly used to improve and accelerate critical parts of the typical enterprise machine learning project, focusing on data ingest/data cleaning, model/dataset version and production deployment.