Last Wednesday, AWS’s CEO Andy Jassy held his traditional keynote at AWS re:Invent, and on the machine learning front, there were several interesting announcements. Here’s a summary of what they were and why you should care…

Amazon SageMaker – What is it?

SageMaker is a fully managed service for implementation, training, automatic hyperparametre tuning, and deployment of machine learning models.

Why should you care?

SageMaker includes a hosted Jupyter environment that doesn’t limit you to a particular machine learning framework -TensorFlow, Caffe, MXnet, CNTK, Keras, Gluon and other major frameworks are all supported. This is in contrast to other Cloud vendors’ fully managed ML offerings, which only offer a single ML framework to work with.

In addition, SageMaker automatically provisions EC2 instances for training and tears them down when the training is complete. This is really handy because up to this point, you had to handle instance provisioning a) manually or b) by implementing your own automation. This annoyance is now a thing of the past.

SageMaker also does automatic hyper-parametre tuning, (no more manual trial-and-error tuning) and model deployment, giving you auto-scaling inference endpoints with very little hassle.

AWS DeepLens – What is it?

DeepLens is a deep learning enabled video camera and associated software toolkit.

Why should you care?

DeepLens includes an onboard graphics processor and over 100 GFLOPS of compute power. What this means in practice is that you can deploy a computer vision model on the device itself and run predictions/inference locally, without a round trip to the Cloud. DeepLens is fully programmable using the AWS Lambda serverless programming model. The models themselves even run as part of a Lambda function. All deep learning frameworks are supported, just like in SageMaker.

Amazon Rekognition Video – What is it?

Rekognition Video does object recognition for video files. Rekognition Video complements the original Rekognition service, which works on image data.

Why should you care?

Object recognition from video previously required you to extract frames from video, convert them to images and then feed them to Rekognition. This process was unwieldy, introducing latency that made it impossible to do near real-time inference. With Rekognition Video, you can do real-time recognition for video, which enables a lot of different use cases. Rekognition Video can detect faces, filter inappropriate content, detect activities and even track people, which is something that other cloud vendors’ object recognition services do not provide out-of-the-box.

Amazon Kinesis Video Streams – What is it?

Kinesis Video Streams is fully managed secure video ingestion and storage service.

Why should you care?

Streaming video to the Cloud is tricky business, typically requiring you to implement your own solution with sufficient protection, scalability and failover mechanisms. It’s a huge hassle, and it’s only a means to an end. A fully managed service that handles all of this is extremely welcome, and in true AWS fashion, it integrates seamlessly with other AWS services.

Amazon Transcribe – What is it?

Amazon Transcribe is machine learning-powered automatic speech recognition and transcription service.

Why should you care?

Transcription typically requires you to hire a transcription service, which may be prohibitively expensive depending on the use case. Amazon Transcription does transcription without manual work, adding in punctuation and, crucially, providing granular timestamps for each uttered word. As with other ready-made AI services, it’ll get better (more accurate) over time and you don’t have to do anything.

Amazon Translate – What is it?

Amazon Translate is a machine learning-powered language translation service.

Why should you care?

Translation services are provided by other Cloud vendors, but until now, AWS hasn’t had their own. Amazon Translate is useful because, as usual, it’s well integrated into other AWS services. It also increases competition in the translation space, which is a win for end users.

Amazon Comprehend – What is it?

Amazon Comprehend is a natural language processing (NLP) service that identifies key phrases, topic, places, people, brands, or events from text. It also does sentiment analysis.

Why should you care?

Entity recognition is, in general, a hard machine learning problem – rolling out your own model takes massive amounts of data, careful algorithm selection and long training times. A ready-made solution allows you to focus on implementing your use case.

If you’d like to know more about these tools, and how best to use them, please contact us here.

Max Pagels

Get in Touch.

Let’s discuss how we can help with your cloud journey. Our experts are standing by to talk about your migration, modernisation, development and skills challenges.

Ilja Summala
Ilja Summala LinkedIn
Ilja’s passion and tech knowledge help customers transform how they manage infrastructure and develop apps in cloud.