BUSINESS TECHNOLOGY MEDIA COMPANY
Companies
Sectors

Five new machine learning tools from AWS

Read time 4min 50sec

Amazon Web Services has just introduced five new machine learning services that it hopes will go some way to solving common problems in the enterprise.

The new services are:

  • Amazon Kendra

This improves the searching of documents inside an enterprise, which, until now, has been almost impossible to search accurately, being fragmented across different departments.

  • Amazon Fraud Detector

This helps businesses identify online identity and payment fraud in real-time, based on the same technology developed for Amazon.com.

  • Amazon CodeGuru

This helps software developers automate code reviews and identify an application’s most expensive lines of code.

  • Amazon Transcribe Medical

This offers healthcare providers highly accurate, real-time speech-to-text transcription.

  • Amazon Augmented AI

The service makes it easier to build and manage human reviews for machine learning applications.

The company unveiled these new products, among others, at its re:Invent conference in Las Vegas last week, partly in response to the fact that there is a paucity of machine learning tools for developers.

According to Matt Wood, VP of AI at AWS, the real value of deep learning will only be unlocked once it’s delivered to a wider community of developers, which will also improve the level of skills in the industry.

“Until today, it wasn’t possible to debug or profile a deep neural network. Trying to do so consumed more compute than training the model itself.”

Joel Minnick, head of AI product marketing at AWS, said it was about three years ago when its customers had begun to ask what machine learning capabilities it could offer them. 

He puts this down to the fact that the cloud had begun to unlock the ability to store huge amounts of data very efficiently, and at the same time had begun to make the ability to process that data more efficient and affordable.

Amazon Kendra

Most enterprises have a host of internal data on their intranets across different department. At present, the information is fragmented, and difficult to search. With its Kendra product, AWS uses machine learning to build an improved enterprise search capability for this storehouse of knowledge. Kendra uses natural language processing to index the words inside documents from different sources. It’s also able to understand the intent of that information, as well as the relationships of the concepts and content between documents.

Amazon Fraud Detector

Building on Amazon Personalize and Amazon Forecast machine learning products, it unveiled Amazon Fraud Detector in Las Vegas. Minnick says Amazon.com ‘has seen just about every type of attempted fraud there is’, and it knew many of its customers were struggling to solve this problem.

Customers provide AWS with their historical transaction data and the rules they use to identify fraud, as well as what kind of fraud they’re attempting to detect. The algorithm will then offer a prediction, in real-time, of which transactions are suspect, through an API call to Fraud Detector.

Amazon CodeGuru

This service automates code review and integrates with AWS CodeCommit, which hosts Git-based repositories. When a developer does a pull request, this automates a call to CodeGuru, which will then revue the code. Minnick says this has been trained using Amazon’s own code reviews, as well as that of the top 10 000 open source projects on GitHub. Once a piece of suspect code has been flagged, it also explains, in English, what the issue might involve, as well as provide suggestions about how to fix it, and links to documentation. It also flags which lines of code it sees as the ‘most expensive’, or, put another way, which lines took the longest to execute. This expensive code is a drain on an organisation’s resources, and identifying these areas can see a rise in efficiency and productivity.

Amazon Transcribe Medical

Minnick says while its Amazon Transcribe speech recognition service has been around since 2017, it has now been improved so that it can be used in the healthcare industry. In the medical field, very specific language is used, and it’s very important that this is understood contextually. Using its Comprehend machine learning tool, the new service has been specifically ‘tuned’ to the language of healthcare, identifying, for example, key phrases, the names of people, places, things, and times, as well as the sentiment.

He says doctors can spend hours of their days transcribing conversations they’ve had with patients, which was inefficient and may contribute to what he called ‘physician burnout’. Amazon Transcribe Medical offers a solution to this.

Amazon Augmented AI

The company said while machine learning can provide highly accurate predictions for a variety of use cases, such as identifying objects in images or extracting text from scanned documents, there are still cases that may require human intervention to clear up any ambiguities. It said human review and machine learning is critical to the success of machine learning systems, but until now it has been hard to manage the human review process.

Amazon Augmented AI is a new service that makes it easier to build and manage human reviews. Developers choose a ‘confidence threshold’ for their application and all predictions with a confidence score below the threshold are automatically sent to human reviewers for validation.

Login with