Abstract: In federated learning (FL), the heterogeneity of data and asynchronous participation of clients have been observed to induce the local client’s model discrepancy with high variance, leading ...