News

Ignite 2019: Improved Azure Cognitive Services Pushes Enterprise AI Forward

• By Joey D'Antoni
• 11/13/2019

During the recent Microsoft Ignite conference, I had a chance to speak with Bharat Sandhu, Director of Product Marketing for Azure AI and Advanced Analytics, and Ria Kapila, Chief of Staff for the Azure AI platform, to discuss the latest AI offerings running on Microsoft Azure.

The ongoing theme is to continue building customer momentum: More than 85 percent of Fortune 100 firms are using Azure AI solutions. Microsoft has made a concerted effort to build and an improve its toolset to make the use of AI solutions more accessible to customers who may not have their own AI engineers on staff.

Customers including Starbucks and Airbus have integrated Azure Cognitive Services from Microsoft into their application stack. This is a set of tools that includes the following services:

• Decision
• Language
• Speech
• Vision
• Web Search

The keynote highlighted uses of the toolset by Spotify and The Atlantic,who scanned their entire archive and integrated Azure Cognitive Services to transition from hard copy to a digital archive can be searched by the public. Specifically, Microsoft has released Form Recognizer and Ink Recognizer which can gather data from journals and return it as JSON. This allows for rich metadata to be extracted and to enrich search models. At Ignite, new speech capabilities allowing for customers to add custom speech models, and translation with automated language detection, were also detailed.

At Ignite, Microsoft also introduced support for virtual network integration with Azure Cognitive Services, as well as an option for container deployment. These features help escalate the services to an enterprise ready production environment. In addition to these infrastructure features, Microsoft is trying to help customers operationalize their machine learning models, using an Machine Learning Operations (MLOps), or DevOps for Machine Learning to help organizations treat their ML models as first class code citizens and to encourage reproducibility of and data of a given model execution. This allows datasets and models to be logged, in order to help build reproduceable models.

Another part of Microsoft's AI journey is integrating AI solutions into their other products. One of these solutions that I use I most frequently is Design Ideas in PowerPoint, which customizes slides in an effort to make there more readable. Cortana integration into Outlook is another example of a feature being built into the Office suite of tools.

Microsoft also introduced a new tool called Power Virtual Agent, which is a bot service that is designed to be implemented by what Microsoft calls "citizen developers" -- business users who build solutions in a low-code environment. The customer example highlighted TruGreen, whose customers could interact with the bots, and more interestingly upload photos of their lawn, which could be analyzed using Azure Cognitive Services, and proper services allocated to the customer without the need for an actual agent.

Microsoft has made eliminating bias and increasing transparency in machine learning models. While many older models that use fairly simple b-tree or regression models, and are fairly easy to interpret, as models move into more advanced neural networks can not necessarily be easily explained to non-technical personnel. So, being able to explain why a given model works the way it does (think of something like a database query execution plan), is highlighted in the Azure ML toolset. In addition to being able to explain models, Microsoft has developed a bias detection tool that developers can apply to trained models. While these toolsets won't capture 100 percent of the bias problems in a model, they do help developers gain awareness of bias, and help them understand.