Abstract:

Federated Learning is a prominent machine learning paradigm which helps tackle data privacy issues by allowing clients tostore their raw data locally and transfer only their local model parameters to an aggregator server to collaboratively train a sharedglobal model. However, federated learning is vulnerable to inference attacks from dishonest aggregators who can infer informationabout clients’ training data from their model parameters. To deal with this issue, most of the proposed schemes in literature eitherrequire a non-colluded server setting, a trusted third-party to compute master secret keys or a secure multiparty computation protocollwhich is still inefficient over multiple iterations of computing an aggregation model.

Description:

.