Computer Vision News - August 2020

Nicola Rieke 39 Best of MIDL 2020 Münster. I moved to the south. So you will get to Italy one day! [ laughs ] Probably! What is the best thing that you have found in Munich that you did not expect? The people! It surprised myself. I have to admit. I came here with a lot of prejudgment. Munich tends to have the cliché that there are a lot of rich and snobbish people. I have to say that when I arrived they were really helpful, very welcoming, very friendly people, very international. I also like Munich very much. Can you tell us about your current research? As I said, I’m still very active in research. Part of my job is staying up to date with recent developments. Probably the best way to do that is doing research myself and being an active part of the research community. One of the topics that I’m working on right now is federated learning. This is a technique that’s been around for a couple of years now, a kind of new learning paradigm that enables collaborative learning without sharing the datasets. Let's say a hospital in East London wants to collaborate with a hospital in West London, but they are not allowed to share patient data. The patient data has to stay secure and cannot be shared. They still want to develop AI together, for example for COVID or for a disease for which a single hospital doesn’t have such data. Federated learning is a way to enable this collaborative learning without sharing the underlying patient dataset. This access to data seems to be one of the problems that this community is going to face in the coming years. Do you have an idea of what the community should do to share data in order to meet the needs of society? There are several answers to that. One of the efforts that has been ongoing for a couple of years is to create centralized data lakes. This is an ongoing effort which takes a lot of organization in terms of analyzing datasets, getting the patient consent, and so on. This is an effort that really fostered a lot of research in the last year. This is one direction that we should definitely follow. However, it is not always possible to share patient data. Patient data is very sensitive, very difficult. We have to develop new techniques to enable the learning without giving direct access to the dataset. If I’m a data donor, and I foster research with that, then I’m more confident to actually give indirect