CodedPaddedFL and CodedSecAgg: Straggler Mitigation and Secure Aggregation in Federated Learning

We present two novel federated learning (FL) schemes that mitigate the effect of straggling devices by introducing redundancy on devices’ data across network. Compared to other in literature, which deal with stragglers or device dropouts ignoring their contribution, proposed do not suffer from client drift problem. The first scheme, CodedPaddedFL, mitigates while retaining privacy level conventional FL. It combines one-time padding for user gradient codes yield straggler resiliency. second CodedSecAgg, provides resiliency and robustness against model inversion attacks is based Shamir’s secret sharing. apply CodedPaddedFL CodedSecAgg a classification For scenario 120 devices, achieves speed-up factor 18 an accuracy 95% MNIST dataset compared Furthermore, it yields similar performance terms latency recently scheme Prakash et al. without shortcoming additional leakage private data. outperforms state-of-the-art secure aggregation LightSecAgg 6.6–18.7 95%.