Here’s how Amazon’s Cloud arm AWS is democratising Machine Learning technology
Amazon at its re:Invent event launched 13 new machine learning capabilities and services that cover all layers in a machine learning stack.Updated: Nov 30, 2018 11:13 IST
While tech giants are busy developing Machine Learning (ML) and Artificial Intelligence (AI) models to deliver enhanced customer experiences, retail giant Amazon’s Cloud arm Amazon Web Services (AWS) is busy dealing with a humongous task -- democratising ML for all and protecting it from “bad actors”.
Some industry watchers say AWS CEO Andy Jassy is overspending on building ML models and services, but for him giving the technology so much time, energy and money is meant to open the doors of companies and developers to the gigantic possibilities ahead once they figure out what exactly is to be done with ML.
Jassy knows what Machine Learning can do when it comes to big and small enterprises and, most importantly, to the differently-abled via enhanced computer vision, brain signal processing, speech recognition systems, voice-based features and other related tools.
However, not many businesses and developers know yet how to fully utilise ML models.
“The problem with ML today is that most companies do not know what they want to do with it. Also, it is equally difficult to find people who know how to build right and ethical ML models and train, tune and deploy those.
“That is why you see us spending so much time, energy and investment in the ML and AL space,” Jassy told the gathering at the company’s annual flagship “ReInvent 2018” conference here.
In the last one year, AWS has released nearly 200 ML services and features.
“We know there is real thrust and hunger from builders to be able to do it more easily, and this is what we want to enable them to do,” said Jassy.
According to him, ML can solve plethora of everyday problems but the real task is to safeguard the algorithms and models from bad actors who have the tendency to harm the world. ALSO READ: Amazon takes on Nvidia, Intel with its own Machine Learning chip
“Even though we have not heard any abuse in our ML models and services so far, we are aware that people will be able to do bad things with these services, the way they do with any technology that can harm the world,” Jassy told a packed house.
One solution, he added, is that ML models and algorithms have to be benchmarked by the industry so that they remain as accurate as possible.
“You have to be clear how you recommend people using those services,” said the AWS CEO as the company announced 13 new ML services and capabilities, including a custom AI chip.
One such ML service is SageMaker that has been given a fresh makeover, to make it easier for customers to build, train and deploy models.
SageMaker “Ground Truth” capability produces high-quality labelled training data; SageMaker RL delivers Cloud’s first managed service for reinforcement learning algorithms and simulators; and a new ML Marketplace offers more than 150 new models and algorithms to developers via SageMaker.
“Over 10,000 customers are currently using Amazon SageMaker. Developers can now sell their ML models and algorithms on AWS Marketplace. It is a great place for developers and customers,” informed Jassy.
Other AI services include Amazon “Textract” that extracts text and data from virtually any scanned document; Amazon “Personalize” and Amazon “Forecast” bring the same technology used by Amazon to developer, with no Machine Learning experience required; and Amazon “Comprehend Medical” tool provides Natural Language Processing (NLP) for medical information.
GE Healthcare is training computer vision models with Amazon SageMaker that are then deployed in its MRI and X-Ray devices.
“By applying reinforcement learning techniques, we are able to reduce the size of our trained models while achieving the right balance between network compression and model accuracy,” said Keith Bigelow, Senior Vice President of Edison Portfolio Strategy, GE Healthcare.
Amazon Comprehend Medical is an NLP service for medical text which uses Machine Learning to extract disease conditions, medications and treatment outcomes from patient notes, clinical trial reports and other electronic health records.
Comprehend Medical requires no Machine Learning expertise, no complicated rules to write, no models to train, and it is continuously improving.
Jassy is confident that irrespective of the challenges ahead, AWS will help democratise ML for all.
“We are not overspending on ML. It is a gigantic area and we are investing right. If you look at the advent of ML and its potential to solve problems, it is huge,” said Jassy.
First Published: Nov 30, 2018 11:12 IST