AWS Re:Invent 2017 – Two Big Time Machine Learning Highlights

Dec 4, 2017 9:33:31 AM

“We want everyday developers (...) to be able to use machine learning much more extensively.” This is Andy Jassy’s mantra targeted at making AWS the company that turns machine learning into a commodity, similar to what the company achieved for IaaS before. Within this context, the following two new offerings stood out of the glut of machine learning and IoT news at Re:Invent 2017.

AWS SageMaker - Machine Learning for the Masses, Finally

AWS SageMaker Delivers turnkey templates for the entire machine learning lifecycle

AWS SageMaker aims to enable developers to enhance any application with machine learning capabilities. ML / AI has remained a niche discipline, due to the absence of developer skills within data center and business disciplines. Enterprise Management Associates is currently working on a research project that aims to quantify the current impact of AI / ML and will show the areas where there’s tremendous potential for ML / AI to improve existing software dramatically.

With AWS SageMaker, Amazon has set out to to change this to put the results of machine learning into the hands of everyone.

Key Value Proposition:

  • Prebuilt Jupyter Notebooks that include the most popular algorythms (K-Means Clustering, Principal Component Analysis, Neural Topic Monitoring, etc.) and machine learning frameworks, such as TensorFlow, Apache MXNet, Caffe2, Pytorch, etc. for immediate deployment.
  • Automatically provide the EC2 instances, storage connections (S3 and Glue), and network interfaces (ENI) needed to run the selected machine learning frameworks.
  • Brings machine learning to the DevOps pipeline by providing artefacts that can be applied within any project context.

Jassy claims that his teams managed to make the algorythms up to “10 times faster than you find anywhere else.” Based on Jassy, this is due to SageMaker algorithms only needing a single pass through the data, where others require multiple. Test runs against Azure and friends will show whether or not these claims are substantiated.

EMA Quick Take: Bringing machine learning algorithms to the masses opens the door for rapid machine learning-driven innovation. Two pizza teams can now quickly experiment with ideas that would have simply been too expensive before. And most importantly, SageMaker could commoditize machine learning and make it available to companies of any size and industries, versus only the largest and most progressive ones. Therefore, every developer, team lead, CIO and CTO should take a look at SageMaker and throw around some crazy ideas for quick POCs. Let’s try this out and see if it is truly as life changingly turn-key as Jassy claims.


AWS DeepLens - A Pre-trained Deep Learning Edge Camera

A network independent machine learning camera running on AWS Greengrass and AWS Lambda

The new AWS DeepLens camera (available in April, 2018) answers the question as to ‘what’s going on with Greengrass.’ Just when I was wondering whether Amazon’s edge compute platform would go anywhere, the company comes out with compelling use cases such as an edge camera with built in deep learning capabilies, running on an Intel Atom-powered Greengrass platform.

Key Value Proposition:

  • Developers can now write ‘serverless’ Lambda functions and run them right on the camera without reliable (or any) connectivity needed.
  • The camera includes a set of pre-trained deep learning models for developers to instantly leverage pretrained deep learning models for standard use cases such as recognition and classification for faces, objects and activities.

Amazon describes deep lense as an edge device with ‘ears, eyes, and a fairly powerful brain.’ While this sounds like the ‘perfect spy,’ this cam is unique in terms of IoT use cases it enables.

EMA Quick Take: Amazon uniquely has 4 ‘secret ingredients’ that could make this cam a ground breaking IoT device and ‘template,’ for similar future devices in different areas:

  1. Lots of experience building cutting edge hardware
  2. The by far most popular ‘serverless’ framework
  3. A ton of AI / ML capabilities
  4. An edge IoT appliance that by now has been around for 12 months

At the same time the new cam doubles as a ‘training lab’ for machine learning, which already is worth the $240 price tag.

Also, check out all the other machine learning and IoT news from AWS Re:Invent. While interesting, these are not unique and mainly aimed to fill out gaps to prevent customers from shopping around.





Torsten Volk

Written by Torsten Volk

With over 15 years of enterprise IT experience, including a two-and-a-half-year stint leading ASG Technologies' cloud business unit, Torsten returns to EMA to help end users and vendors leverage the opportunities presented by today's hybrid cloud and software-defined infrastructure environments in combination with advanced machine learning. Torsten specializes in topics that lead the way from hybrid cloud and the software-defined data center (SDDC) toward a business-defined concept of enterprise IT. Torsten spearheads research projects on hybrid cloud and machine learning combined with an application- and service-centric approach to hyperconverged infrastructure, capacity planning, intelligent workload placement, public cloud, open source frameworks, containers and hyperscale computing.

  • There are no suggestions because the search field is empty.

Lists by Topic

see all

Posts by Topic

see all

Recent Posts