Decentralized Learning (DL) allows better privacy-preserving machine learning for edge computing systems. There is a need to design novel DL protocols that are robust to privacy threats, and deal with heterogeneity of decentralized DL architectures.