Build Ahead of next week's Build, Microsoft's annual developer shindig, the US tech giant is continuing its efforts to show it knows how to play nicely with others – as it preps a slew of updates to its Azure Cognitive Services.
These details emerged today just as Azure was busy recovering from a global DNS-related outage, we must note. And so, on with the details of the new shiny shiny:
It looks like you're trying to make a decision. Do you want some help with that?
While trumpeting the 1.3 million developers building apps on its Azure Cognitive Services platform, Microsoft also announced the arrival of the distinctly creepy-sounding service: "Decision". Including the even creepier-sounding "Personalizer", the technology picks up knowledge from experience and then offers recommendations to help meatsacks make decisions.
The tech is based on reinforcement learning and, according Lance Olson, an AI boffin at Microsoft, doesn’t need data scientist-level skills to be made to work.
The AI gang is also flinging out public previews of tech that can learn to read handwriting, in the form of Ink Recognizer, and Form Recognizer which is, er, able to recognise forms (including the likes of key value pairs and tables.)
For the former, Microsoft already has some recognition smarts in Azure Cognitive Services in the form of Computer Vision but, in the world of Redmond, you can never have too many pastings from the AI brush.
Microsoft is also adding conversation transcription capabilities, handy for dodging the "who wants to be the scribe?" question in meetings, and tweaks to speech, vision and languages.
Finally, the company is making cognitive search generally available, inflicting AI on Azure Search. The gang is also adding previewing tech to store insights gleaned by AI from cognitive search; handy for developers keen to create visualisations and machine learning models.
ML Interoperability and Acceleration
Keen to get data scientists on-side, Microsoft boasted that it is now an active contributor to the MLflow project, an open source platform for the machine learning lifecycle. MLflow works with pretty much any ML library or language (according to the maintainers) and will run the same way on any cloud.
So not just Azure then.
Currently at version 0.91 (released on 21 April), MLflow consists of three components. There is Tracking, an API and user interface for logging data like parameters when running ML code as well as visualising the results. The second component is the packaging format Projects, and the third is Models, a format for sending models to deployment tools.
Microsoft is also a fan of the Open Neural Network Exchange (ONNX) runtime to create deployment models that work across multiple platforms. To that end it has made available support for NVIDIA TensorRT and Intel nGraph chips for high-speed inferencing.
Naturally, MLOps (DevOps for Machine Learning) also receives a nod because heck, the industry just doesn't have enough acronyms, initialisms or portmanteaus to go around, right?
Essentially a Continuous Integration and Continuous Delivery (CI/CD) pipeline for AI and ML, Azure DevOps is used along with Azure ML Services for model training, management and operationalisation as well as the other DevOps tasks, like unit testing.
Finally, on the AI/ML front, new ML notebooks are there for code-first development, while those reluctant to get their hands dirty are rewarded with Azure Machine Learning's "automated machine learning experience". For those sat between the two stools, there is a visual drag and drop interface to build models without code.
It's all very noble, but we’d wager it will be hard to avoid some level of coding once things start getting complicated.
The third horseman of the Hypocalypse: Blockchain
The Azure Blockchain Workbench - infrastructure scaffolding for building applications using the ledger technology - has been in preview for a while now. Today, Microsoft added a preview of the Azure Blockchain Service to its line-up with a view to simplifying the formation, management and governance of consortium blockchain networks.
Consortium blockchains, for those who glazed over at the first mention of the 'B' word, sit between open and private blockchains, having a controlled user group across different organisations.
The service integrates with Azure Active Directory to allow administrators to manage users and an interface in the increasingly crowded Azure Portal to keep an eye on health and activity within the permissioned blockchain network. The whole thing can be kicked off with a few mouse-clicks.
And the chosen ledger for Microsoft’s shiny new service? J.P. Morgan's Quorum, built on Ethereum. ®