Gradient Disaggregation: Breaking Privacy in Federated Learning by
           Reconstructing the User Participant Matrix

     Maximilian Lam 1 Gu-Yeon Wei 1 David Brooks 1 Vijay Janapa Reddi 1 Michael Mitzenmacher 1

             Abstract               or wearables, and maintaining the integrity of their data is a
    We show that aggregated model updates in fed-      critical issue. Already, federated learning has been deployed
    erated learning may be insecure. An untrusted      by many maj ...