Technology Fridays: About Watson’s New Deep Learning as a Service Vision

Image for post
Image for post

Welcome to Technology Fridays! Today, I would like to deep dive into one of the major announcements at IBM’s Think conference that took place this week. At the event, big blue unveiled a new suite of artificial intelligence infrastructure and platform services under the catchy name of Deep Learning as a Service(DLaaS). The new stack extends the capabilities of Watson Studio to streamline the implementation and operationalization of deep learning solutions across different stacks.

Beyond the marketing hype, IBM’s DLaaS vision has some very interesting ideas that are worth exploring. From the market perspective, the release is an attempt to bridge the gap with the machine learning cloud services provided by platforms such as Microsoft Azure or Google Cloud that have been taking a leadership position as enablers of cloud deep learning solutions. However, IBM’s DLaaS is not merely a replication of deep learning capabilities existing in other platforms and it provides some very unique technology components assembled in a cohesive story.

I mentioned before that IBM’s DLaaS enhancements live within Watson Studio. The main contribution of the new stack is a consistent cloud runtime for applications developed using different deep learning frameworks such as TensorFlow, PyTorch, Keras or IBM’s favorite Caffe.

Image for post
Image for post

The DLaaS stack provides a series of tools and services that streamline different aspects of the lifecycle of deep learning applications. As a matter of fact, if you can visualize the typical lifecycle of a deep learning application from experimentation to optimization, you will find different DLaaS tools and services on each stage of the cycle. Let’s explore some of the important elements of the IBM DLaaS suite.

Experiment Assistant

IBM DLaaS Experiment Assistant are a series of tools and runtime components that automate the workflows for training and evaluating deep neural networks. Data scientists can use the Experiment Assistant to configure, execute and monitor training workflows for deep learning models without having to be concerned about the underying infrastructure.

Image for post
Image for post

Neural Network Modeler

Another important addition to IBM DLaaS is the Neural Network Modeler. This new tool enables the implementation of deep learning programs by providing an intuitive drag-and-drop interface that enables a non-programmer to speed up the model-building process by visually selecting, configuring, designing and auto-coding their neural network using the most popular deep learning frameworks.

Image for post
Image for post

Fabric for Deep Learning

As part of the release of Watson DLaaS, IBM is open sourcing the core runtime of the platform. Named the Fabric for Deep Learning (pronouncedFfDL), the new open source package provides a scalable, resilient, and fault-tolerant runtime for the execution of deep learning programs. The Fabric for Deep Learning leverages technologies such as Kubernetes and Uber’s Horovod to provide a distributed environment for the execution of deep learning models.

Competition?

IBM DLaaS can be seen as a series of addition to Watson Studio rather than a standalone release. From that perspective, the new stack is a competitive alternative to platforms such as Azure ML, AWS SageMaker or Google Cloud ML. If many thought that IBM was faling behind the other cloud incumbents in the race to become the preferred home for deep learning applications, the release of the DLaaS suite signals to the market that IBM has a very unique and innovative vision about the space .

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store