Federated learning (FL) has emerged as a popular paradigm for distributed machine learning over decentralized data. Data generated by FL clients is prone to noises. While the impact of data noise on centralized learning (CL) is well understood, there …