Abstract Federated learning (FL) is a promising paradigm for collaborative, privacy-preserving training of machine learning (ML) models on decentralized, private data. However, significant challenges limit the broader adoption of FL. First, existing FL workflows are often tailored to specific use cases, requiring extensive manual setup and customization of execution environments. Second, establishing collaborations and matching partners is complicated by the diversity of organizational and collaboration goals, data properties, and data distributions. Third, tracking the provenance of the collaboration process and the artifacts created and used during collaboration is challenging, impeding accountability, transparency, and debugability. Our goal is to address these barriers by lowering the technical expertise required from collaborators. We contribute mechanisms for flexible collaboration composition and creation, automated collaborator matching, and collaboration and artifact provenance.
Schlegel et al. (Sat,) studied this question.