The success of modern machine learning is based on access to rich, diverse, and, above all,large data sets. However, getting access to large datasets can be a challenge in many domains, one of which is the medical domain. Different institutions, such as hospitals, medical centers, and pharmaceutical companies, often own the data, and there are straight privacy and regulatory constraints when sharing such data. Moreover, medical data is sometimes collected by IoT devices with their limited inherent communication and privacy constraints. Hence, despite the benefits of machine learning in such distributed data settings, it is essential that the training be done locally without sharing any data, but by resorting to distributed optimization solutions, such as Federated Learning. Federated learning for medical data is now in its infancy, while medical data has many unique challenges, e.g., in terms of data-owners, regulatory concerns, data diversity, algorithmic fairness, and biases. The goal of this special track is to focus on recent advancements around federated learning in such medical settings.

The topics of this special track will include but will not be limited to the following:
1.Federated transfer learning for medical data sources
2.Privacy-preserving techniques for federated learning
3.Architectures and protocols for federated learning on medical data sources
4.Explainable federated learning medical data sources
5.Federated learning for univariate and multivariate medical time series
6.Federated learning for time series nowcasting and forecasting
7.Handling data diversity in federated learning architectures
8.Federated learning for medical IoT
9.Personalization in federated learning
10.Heterogeneous and unbalanced (non-IID) medical data

Organizers: Sindri Magnússon, Ioanna Miliou, Panagiotis Papapetrou

More details of this Special track please see: