Federated learning is a technique which allows training machine learning models in a distributed way without transferring the data from its source. It has thus potential applications in analysis of medical data, where privacy and security issues are of great importance. Although there are examples of using federated approaches to analysis of medical data, there is still need for research in this area and for experiments in real distributed environments.
The goal of the thesis will be to apply federated learning techniques to selected problems in area of medical data analysis in collaboration with other Sano research teams. Examples of the data are coming from the various areas of research at Sano, including but limited to:

• Medical image classification and segmentation,
• Individual patient data coming from mobile applications and wearable devices,
• Genomics data from next generation sequencing,
• Training of medical robots.

The analysis will be performed using distributed computing frameworks such as Flower, FedML or Nvflare, using distributed computing infrastructures such as PL-Grid, EuroHPC or public cloud services. The main challenge will be to evaluate how the large scale of datasets influences the machine learning process, and how to optimize it for such scenarios.
In addition to evaluation of the learning process, the goal will be also to evaluate the performance of distributed computing environment. Various metrics related to the distribution of data, its granularity and partitioning will be investigated, to understand their impacts on the both the efficiency of the learning process and the performance of the infrastructure. Further study will include also possible attacks and security of the developed solution. Other types of data and machine learning tasks can be considered for comparison as well.