Microsoft is continuing to look for ways to make machine-learning technology easier to use. In 2018, Microsoft bought Lobe, a San Francisco-based startup that made a platform for building, training and shipping custom deep-learning models. This week, Microsoft made some of Lobe's technology publicly available.
On October 26, available a public preview of a Lobe app for training machine-learning models. Available for both Windows and Mac, the Lobe app is free and designed to enable people with no data science experience to import images into the app and label them to create a machine learning dataset.
According to Microsoft, "Lobe automatically selects the right machine learning architecture and starts training without any setup or configuration." Users can evaluate the model's strengths and weaknesses via real-time visual results and modify the model to boost performance. Once they're done, users can export their models to a variety of industry-standard formats and ship it on any platform they chose, according to Microsoft's Lobe web page.
Officials said while Lobe currently supports image classification, it will be expanding to other model and data type templates in the future. On the "coming soon" list are object detection, which will enable location of an object inside of an image and data classification, which will label data in a table based on its content.
In an interesting twist, Microsoft is positioning Lobe as better suited for those worried about privacy because they don't need to upload their data to the cloud to use it. Officials said the app, which trains and stores data locally, complements Microsoft's various Azure AI services.