Abstract: In this paper, we develop a unified framework for joint model pruning and link scheduling in decentralized learning systems to improve model consistency and communication efficiency. The ...