FedCast: Personalized predictions without sharing data

What is the project about?

FedCast is an open-source toolkit for personalized federated learning with timeseries data. Whether in healthcare for individual therapy planning, in energy supply for consumption optimization, or in industry for predictive machine maintenance, accurate and personalized predictions are crucial. However, previous approaches often require the central storage of sensitive data, which not only poses a significant risk to data privacy but also undermines user trust in such technologies. FedCast enables the development of personalized prediction models directly on users' devices without sharing sensitive information with third parties or storing it centrally.

Which audience does the project address?

The target audience for this project consists of researchers, developers, and data scientists in the fields of machine learning and timeseries analysis who want to use or evaluate federated learning in their projects.

What is to be achieved?

The goal is to provide a robust, adaptable toolkit that can be used to quickly and easily develop applications for personalized federated learning. It should be possible to combine different algorithms and methods.