Unlocking Inaccessible Data Sources with Active Learning at Utility Company Stadtwerke München
Tuesday, May 12, 2020
Important information is often buried in graphical plans, hand written notes and other sources which are almost impossible to access at scale, even with the help of computer vision or optical character recognition. With Active Learning, you get a huge reduction in human effort in return for a somewhat lower precision. Sarah and Michael demonstrate how they successfully applied Active Learning to an archive consisting of partly handwritten plans which can only be interpreted with expert knowledge on company specifics. They discuss which algorithms from deep learning and classical ensemble learning can be used together with this technique.