News Categories

re:Invent 2017: DeepLens and SageMaker make machine learning a reality for all developers

By Michael Low - on 30 Nov 2017, 7:00am

re:Invent 2017: DeepLens and SageMaker make machine learning a reality for all developers

AWS DeepLens, the world's first deep learning-enabled video camera. <br>Image source: Amazon Web Services.

If you're wondering where Amazon Web Services (AWS) is heading in terms of machine learning (ML), it's to make it easier and accessible for every developer and data scientist to build, train, and deploy ML models.

Amazon SageMaker includes hosted Jupyter notebooks for visualizing training data stored in S3.

Amazon SageMaker, which is generally available today, is a fully managed service that takes away the complexity of ML implementation – e.g. compute power, retrain time, hyper-parameter tuning, specialized skills – by providing 10 commonly-used algorithms, broad framework support, as well as one-click training at petabyte scale, and one-click tuning (via hyper-parameter optimization) for highest possible accuracy.

The ensuing ML model can then be deployed on an auto-scaling cluster of EC2 instances across multiple availability zones. There are also one-click inference, native A/B testing support, and secure HTTPS endpoint for high throughput and low latency predictions.

But that's not all, as AWS also revealed the world's first deep learning-enabled wireless video camera, DeepLens. Aside from a HD video camera, the custom-designed hardware is capable of running over 100 billion deep learning operations per second. As it comes complete with sample projects and pre-trained models, developers can get started in less than 10 minutes, regardless of skill levels. It's also fully programmable with Lambda functions, and can be integrated with SageMaker to run and deploy trained ML models on the device in real time. 

Dr. Matt Wood, General Manager, Artificial Intelligence, AWS demonstrated a music recommendation engine, where the DeepLens camera detects album covers and facial expressions to suggest recommendations based on the user's positive or negative reaction.

AWS DeepLens will retail for US$249 (~RM1,018) on in 2018.