Machine learning (ML) remains a major growth area for Amazon Web Services (AWS), with the group launching 13 new ML services and capabilities at AWS re:Invent in Las Vegas this week.
This includes a custom chip to speed up machine learning training and an autonomous toy race car for developers to practise reinforcement learning.
"If you look at the last couple of years, we hardly get through conversations with companies without talking in some significant part about machine learning," AWS CEO Andy Jassy told journalists at a press briefing at the event.
"I think we are entering this golden age of what is going to be possible in applications, and in five years from now or 10 years from now, I think virtually every application will have machine learning and AI in it."
Last year's re:Invent also had a big focus on artificial intelligence (AI) and ML, when the company announced Amazon SageMaker, a managed service for developers and data scientists to quickly build, train, deploy and manage their own ML models.
This year, the company continued to build on the success of SageMaker, which Jassy said is being used by over 10 000 customers after one year in existence.
"What really drives what we build is what our customers tell us matters and we have a lot of customers talking about machine learning."
Jassy made a slew of announcements during his keynote speech, including some new AI services.
He covered Amazon Textract, which extracts text and data from virtually any scanned document; Amazon Comprehend Medical, which provides natural language processing for medical information; and Amazon Personalize and Amazon Forecast, which provide customised personalisation, recommendations and forecasts using the same technology used by parent company Amazon.com.
Jassy said what AWS built in terms of ML and AI over the past year was "very much a continuation from what we talked about last year".
"We just have so many customers who want to be able to use machine learning and they need help in every layer of that machine learning stack. They need more powerful compute instances that allow them to scale their models better; they need a way to actually support all of the frameworks they want to do training with.
"Even though the incremental progress each year is astounding, it's still pretty early for most companies in terms of knowing what they want to do with machine learning and having the people who know how to build machine learning models and train them, tune them and deploy them. And then also the capabilities being there so that you can do it more easily," he explained.
He said that is why AWS has been spending so much time and investment, and has made so many releases in the ML and AI space in the last few years.
"Just in the last year, we have released 200 significant services and features in machine learning and AI, and that is just because we know there is a real thirst and hunger from builders to be able to do it more easily.
"You can expect that we are not close to being done. I think that some people think we are overinvesting in this area, but I don't think so. I think this is going to be a gigantic area. I think that virtually every application over time will be using these machine learning services."
New capabilities
A lot of the ML services and capabilities announced this year are interesting but a particularly fun one is the AWS DeepRacer, a $399 (R5 456) autonomous toy car, which is one-eighteenth the size of a real race car, which aims to help developers get rolling with machine learning.
The model race car is driven using reinforcement learning models trained using Amazon SageMaker. It includes a 3D simulation environment powered by AWS RoboMaker, also announced at re:Invent, which makes it easy for developers to develop, test and deploy robotics applications and build intelligent robotics functions using cloud services.
"Reinforcement learning is enabling innovation in machine learning and robotics," said Brad Porter, VP and distinguished engineer of Amazon Robotics.
"We're excited Amazon SageMaker is making it easier to try reinforcement learning techniques with real-world applications, and we're already experimenting with ways to use it for robotic applications. For instance, earlier this year, we showed a robot that was able to play beer pong using some of these techniques and we're excited to continue to explore these opportunities in partnership collaboration with AWS," Porter added.
AWS unveiled several Amazon SageMaker features, including low-cost, automatic data labelling and reinforcement learning, and also announced a high-performance ML inference chip, AWS Inferentia, which was custom-designed by AWS and will be available in 2019.
Jassy also announced a new AWS Marketplace for machine learning. He said ML is moving quickly, with new models and algorithms from academia and industry appearing virtually every week.
Amazon SageMaker includes some of the most popular models and algorithms built-in, but to make sure developers continue to have access to the broadest set of capabilities, the new ML marketplace includes over 150 algorithms and models, with more coming every day, which can be deployed directly to Amazon SageMaker.
Share