Computer Vision News - January 2022

46 Exclusive Interview papers in Nature Machine Intelligence on this topic in the last two years. Who are your users? Our users are mostly radiologists who work with medical imaging. The tools or algorithms we develop either simplify their tasks or allow them to make a diagnosis directly from the images. For example, measuring the size of someone’s liver, or whether someone is likely to have a certain disease or not. Feedback from the radiologists tells us whether these difficult to get data for medical research here. My colleagues, including Georgios Kaissis, and I are working on privacy preserving machine learning. This is a type of machine learning where you use data either in an encrypted fashion, or by leaving it where it is and taking the algorithm to it. This is what people call federated learning. We are also using differential privacy to make sure no one can reverse-engineer these models once we have constructed them. This gives mathematical guarantees on preserving privacy in a machine learning scenario. We have had a couple of This is what people call federated learning!