Machine Learning Foundation: An Architectural Overview

Posted by Frederic Koppenberg on September 22, 2017

More by this author

Never before have we had access to so much information – and we’ve never had the opportunities to streamline and leverage that data the way we can with machine learning.

Here at SAP we’re right at that intersection of innovation, which is why we’ve developed the SAP Leonardo Machine Learning Foundation (or ML Foundation for short). Margaret Laffan recently gave us a and today we want to tell you more about the services it offers.

Leveraging Cutting-Edge Technology

We already know that machine learning uses complex algorithms instead of programming code to detect patterns in large, unstructured samples of data and make applications intelligent.

Whereas more traditional machine learning services rely on CPUs for this process, we partnered with NVIDIA so that the ML Foundation utilizes cutting-edge Graphics-Processing Units – that is, GPUs instead of CPUs.

What’s more, because the ML Foundation is a cloud offering, its set up is straightforward, scalable, and more efficient than on-premises solutions. It’s already tightly integrated with the SAP Cloud Platform, making the development of other apps in SCP with ML Foundation easy.

ML Foundation – Endless Use Case Applications

Let’s take a look at a use case for how the ML Foundation works for business, with the example of the jeweler and jewelry repair.

You can imagine the challenge. A small part of an intricate piece of jewelry needs fixing or replaced, but to find the specific part in an existing product database would require sifting through thousands of products. This process is both cumbersome and timely.

To solve this problem with ML Foundation, we developed an application with the simple architecture of a database, a REST client, and an UI5 based user interface on the SAP Cloud Platform. In this solution, the existing product catalogue – the images of jewelry – is saved in the database as an image blob. A simple REST client was developed to orchestrate calls to the database, calls to the ML Foundation and to expose OData services that are consumed by the UI5 application.

When an image of broken jewelry is sent to the jeweler, it can be uploaded using the user interface. The Image Feature Extraction Service uniquely built into the ML Foundation then calculates a feature vector of the image that captures its distinct characteristics.

Then another service of the ML Foundation –  Similarity Scoring – compares the feature vector with the existing product catalog and ranks the products according to their similarity to the broken part. The correct match can then be selected from the user interface and the ordering process for the necessary spare part can be initiated.

 

Instead of a tedious manual process, the jeweler has the immediate ability to find the part and deliver the repair quickly and efficiently, giving the business a competitive edge.

Advanced Options for Developers

As you can see, the possible business solutions are endless, and the prospects for developers are equally as exciting. For developers, the ML Foundation offers pre-trained services that provide readily consumable models.

These can be used as a web service by calling simple REST APIs. These RESTful web services are used by SAP´s internal developers for products such as S4HANA, Fieldglass, and more, but they can also be used by customers for their own projects. In the future, developers will have even more advanced options to use ML Foundation.

In the meantime, stay tuned for   at the end of September! For more information on how to get started with SAP Leonardo Machine Learning Foundation, have a look on Hanna’s Blog Article and the SAP Leonardo Machine Learning Webpage.

VN:F [1.9.22_1171]
Average User Rating
Rating: 3.8/5 (11 votes cast)
Machine Learning Foundation: An Architectural Overview, 3.8 out of 5 based on 11 ratings

4086 Views