Efficient Federated Learning over Multiple Access Channel with Differential Privacy Constraints

In this paper, the problem of federated learning (FL) through digital\ncommunication between clients and a parameter server (PS) over a multiple\naccess channel (MAC), also subject to differential privacy (DP) constraints, is\nstudied. More precisely, we consider the setting in which clients in a\ncentralized network are prompted to train a machine learning model using their\nlocal datasets. The information exchange between the clients and the PS takes\nplaces over a MAC channel and must also preserve the DP of the local datasets.\nAccordingly, the objective of the clients is to minimize the training loss\nsubject to (i) rate constraints for reliable communication over the MAC and\n(ii) DP constraint over the local datasets. For this optimization scenario, we\nproposed a novel consensus scheme in which digital distributed stochastic\ngradient descent (D-DSGD) is performed by each client. To preserve DP, a\ndigital artificial noise is also added by the users to the locally quantized\ngradients. The performance of the scheme is evaluated in terms of the\nconvergence rate and DP level for a given MAC capacity. The performance is\noptimized over the choice of the quantization levels and the artificial noise\nparameters. Numerical evaluations are presented to validate the performance of\nthe proposed scheme.\n