Fraunhofer Institute for Digital Medicine MEVIS
Grand Challenge is a well-known platform for end-to-end development of machine learning solutions in biomedical imaging. The platform is open source, developed largely by the Diagnostic Image Analysis Group at the Radboud University Medical Center Nijmegen, and the main installation runs on AWS, with currently nearly 100,000 registered users, several hundred registered challenges, and many ready-to-run AI algorithms.
On the one hand, the use of public cloud technologies facilitates the scalable public deployment, which allows many concurrent users to run algorithms on cloud machines on data uploaded into the platform. On the other hand, institutional regulations may not allow the use of the platform with local research datasets, e.g., if they must not be uploaded to commercial cloud platforms. Such issues motivate a more decentralized approach.
Fraunhofer MEVIS is running a GPU cluster on premise that serves as so-called “edge cloud”. This means that we can offer similar services with similar technologies as the public clouds, but the data stays on our local machines. Based on this edge cloud technology, Fraunhofer MEVIS wants to provide a Grand Challenge node to the research community that allows to share data or algorithms with each other while still keeping them in a German non-profit organization's datacenter.
The installation shall not be totally isolated, however, we plan to integrate with the main Grand Challenge platform where it makes sense. For instance, it should be possible for algorithm owners who published their research results on one of the two platforms to share it with the other community by explicitly sending their algorithm to the other node. We also do not want to host another public database of MIC challenges, for which Grand Challenge is famous and should stay the address to refer to, but challenges hosted on our node should automatically be registered with the main Grand Challenge site as well.