- Authoring: Zero-setup hosted Jupyter notebook interfaces for data exploration and cleaning. These can be run on general instances or GPU-powered instances.
- Model Training: A distributed service for model building, training, validation. You can either use the built-in common unsupervised and supervised learning algorithms and frameworks, or create your own training using Docker containers. To support faster model building, the training can be scaled to hundreds of instances. S3 is used to read the training data and S3 is used to store the model artifacts. The model artifacts refer to the data dependent parameters of your model, not the code that allows for inferences. This separation of concerns allows you to easily deploy Amazon SageMaker-trained models to other platforms, such as IoT devices.
- Model Hosting: A model hosting platform with HTTPs endpoints that allow you to invoke your models and get real-time inferences. These endpoints are scalable to handle traffic and allow you A/B test multiple models simultaneously. These endpoints can be built using the SDK, or you can create your own configurations using Docker images.
Hunt stated that each component can be used in its own right or in combination. SageMaker is currently only available in the AWS Northern Virginia, Oregon and Ohio regions. You can find more information, including pricing, here. AWS DeepLens Jassy also revealed a new edge device called AWS DeepLens. This allows developers to build machine learning models that incorporate real-world objects such as faces or scenes. DeepLens is a Wi-Fi-enabled portable video camera that features a 4MP lens, 1080P support, and a 2-D microphone array. It is powered by an Intel Atom processor that has 8GB RAM and 16GB expandable memory. It also has USB 2.0, HDMI, and Micro SD ports. It runs Ubuntu 16.04 and includes Greengrass Core. There are also pre-trained machine learning models that can detect and recognize images. [Click on the image to see a larger view.] The specifications of the DeepLens camera. Source: AWS. DeepLens camera specs.