D-Cliques: Compensating for Data Heterogeneity with Topology in Decentralized Federated Learning | IEEE Conference Publication | IEEE Xplore