Enhancing federated learning privacy using Sketching
Enhancing the Privacy of Federated Learning with Sketching
AI summary
The research proposes using Sketching algorithms to enhance federated learning privacy; preliminary results of a prototype on three representative learning models show that it can provide strong privacy guarantees without sacrificing performance or accuracy. Although federated learning keeps data local, shared model updates may still leak private information such as weight and height; existing methods often trade off among performance (especially communication cost), accuracy, or privacy.