Abstract

https://github.com/carlynlee/credo-api-tools/tree/playground/ This project aims to design an architecture for applying Federated Learning (FL) to the analysis of cosmic ray event data collected through the Cosmic Ray Extremely Distributed Observatory (CREDO) project. CREDO gathers data from a global network of smartphones, enabling a large-scale citizen science initiative to study cosmic rays. The traditional approach of centralizing this data raises concerns about privacy and data ownership. We explore as a potential solution, allowing decentralized training of machine learning models directly on participants' devices. We explore how FL can accommodate CREDO's requirements, such as handling heterogeneous data and robust model aggregation. We present our initial steps in designing an FL architecture tailored to the needs of the CREDO project, outline our architectural proposal, and discussing the potential benefits and the technical challenges. This sets the stage for future implementation and evaluation, aiming to enhance accountability and collaboration in scientific data analysis.