The paper introduces the Fed-PLT algorithm for federated learning, focusing on reducing communication rounds and enhancing privacy through local training. It addresses challenges of expensive communications and privacy preservation. The algorithm allows partial participation and various local training solvers without compromising accuracy. Differential privacy bounds are derived, and the algorithm's effectiveness is compared to alternative techniques.
Til et andet sprog
fra kildeindhold
arxiv.org
Vigtigste indsigter udtrukket fra
by Nicola Basti... kl. arxiv.org 03-27-2024
https://arxiv.org/pdf/2403.17572.pdfDybere Forespørgsler