Weighted distributed differential privacy ERM: Convex and non-convex

Distributed machine learning allows different parties to learn a single model over all data sets without disclosing their own data. In this paper, we propose weighted distributed differentially private (WD-DP) empirical risk minimization (ERM) method train in setting, considering weights of clients. For the first time, theoretically analyze benefits brought by paradigm learning. Our advances state-of-the-art ERM methods setting. By detailed theoretical analysis, show that noise bound and excess can be improved held multiple parties. Additionally, some situations, constraint: strongly convexity loss function is not easy achieve, so generalize our condition restricted convex but satisfies Polyak-Łojasiewicz condition. Experiments on real more reliable improve performance ERM, especially case scales clients are uneven. Moreover, it an attractive result achieves almost same experimental results as previous centralized methods.