At Build 2020 today, Eric Boyd, CVP of Microsoft Azure AI, reaffirmed the company’s beliefs regarding an increasing need in today’s world to build AI systems that “behave fairly” and comply with privacy regulations. In order to advance this Responsible ML phenomenon, at its Build event today, the tech giant has unveiled new innovations in Azure Machine Learning.
The primary goal through these enhancements is to help developers better understand, control, and protect their developed models. As such, Fairlearn – a toolkit which assesses and improves the fairness of AI systems – is planned to be integrated with Azure ML next month. Moreover, the source code for WhiteNoise – a toolkit for differential privacy – has been made available on GitHub. The latter was developed in collaboration with Harvard’s Institute for Quantitative School Science.
Meanwhile, the Scandinavian Airlines (SAS) is also planning to create AI solutions using Azure Machine Learning including a system that optimizes fresh food stocking for in-flight purchases and a system that forecasts ticket sales. The airline service has also developed such services that leverage Azure ML in the past. As far as the “control” aspect is concerned, new features in the cloud-based platform include “fine-grained traceability, lineage, and access control of data, models”, and more. New workflow documentation capabilities to enforce accountability have been introduced as well.
With regards to enhancing its role in offering machine teaching capabilities, Microsoft has announced a new machine teaching service in public preview, that helps “create and optimize intelligence for industrial control systems”. Dubbed Project Bonsai, the service enables subject-matter experts to add intelligence to AI systems without necessarily requiring them to have a background in AI.
Project Moab – an 3D-printable, open-source balancing robot that assists with the creation of real-world autonomous control systems using Project Bonsai has also been unveiled. Purchase availability for the robot is planned to be announced further along this year.