Jaggi, MartinHartley, Mary-AnneBerdoz, Frédéric2021-07-092021-07-092021-07-092021https://infoscience.epfl.ch/handle/20.500.14299/179843Federated and decentralized learning have become key building blocks for privacy-preserving machine learning. Participation in these opaque federations may be better incentivized by transparent communication of each user's contribution. For real-world applications with large numbers of heterogeneous participants, quantifying these contributions according to their impact on model quality remains challenging. We discuss the applicability various contribution measures with a particular focus on the personalized learning setting, where each participant has their own learning objective.Collaborative LearningFederated LearningDecentralized LearningPeer-to-Peer LearningContribution MeasureIncentivizationPersonalizationstudent work::semester or other student projects