Microsoft wants to make it easier for anyone to build machine learning systems through the release of a newly revamped app that caters to people with little or no coding experience.
Microsoft Lobe, which is available in public preview this week, allows users to build AI-powered visual recognition systems by simply uploading sets of labeled images for training. Its current customers range from marketers looking to identify optimal images for social media ad performance to nonprofits attempting to map marine resources by scanning social posts of people whale watching.
Microsoft is one of many companies rolling out tools to simplify and automate the process of coding basic machine learning systems. A lack of technical talent and daunting complexity has hampered AI adoption among some businesses. Google, IBM and Amazon have all released offerings that similarly make AI development more accessible with varying degrees of technical understanding needed.
Lobe boasts a much less technical interface than even many of these tools, however. The program can train on as few as five images (though more are better for accuracy), then automatically selects the right architecture for the data. Users can also tweak results as the model trains to improve accuracy.
“Lobe is taking what is a sophisticated and complex piece of technology and making it actively fun,” Bill Barnes, manager for Lobe, said in a blog post announcing the release. “It fills [people] with confidence that they can actually use machine learning. And when you have confidence you become more creative and start looking around and asking ‘What other stuff can I do with this?’”
In a test of the app, we were able to train a model to fairly accurately distinguish between packs of Starburst and Skittles in around 5 minutes with 15 photos of each.
But while more wide-ranging platforms like Amazon Rekognition and Microsoft’s computer vision API can identify images with more dexterity, Lobe is best for use cases specific to the needs of a particular project.
For instance, Ansira-owned automative marketing agency Sincro uses the tool to distinguish photos of vehicles as they appear in a sales lot from generic stock images, which don’t perform as well in social ads, for local car dealerships, according to Microsoft’s announcement.
The company is planning to eventually add more options for other simple types of machine learning functions, including object detection and data classification. Microsoft acquired Lobe in 2018 and has been incubating it into its current version since then. It’s now available for download on Windows and Mac.