Deep learning platform aims to speed up developments

One year ago, the would-be founders of the ETH spin-off Mirage Technologies won first prize at the Hackzurich programming competition, as explained in a press release issued by the Swiss Federal Institute of Technology in Zurich (ETH). The application developed by Igor Susmelj and his fellow students was named Deepbusters and was able to expose fake videos published online. However, it quickly became apparent that this kind of platform was not going to make much money.

Things should be different for the new product developed by Susmelj and his colleague Heiki Riesenkampf, an ETH computer scientist. The deep learning platform from Mirage Technologies provides companies with pre-programmed and trained models for the development of products. Individual “families” of these models are trained in different issues, for example the recognition of objects. Moreover, the models are capable of further learning and can be “fed” with additional data for specific tasks, the press release explains. “As we don’t have to train our model from scratch, we need much less data”, Susmelj commented.

“Many companies want to try out new technologies”, Susmelj explains. Mirage Technologies has made the basic features of the deep learning platform available free of charge for now. This aims to build up a client base that will in future be prepared to pay for additional products and services.