E-Tree Learning: A Novel Decentralized Model Learning Framework for Edge AI

Traditionally, AI models are trained on the central cloud with data collected from end devices. This leads to high communication cost, long response time and privacy concerns. Recently Edge empowered AI, namely has been proposed support model learning deployment at network edge closer sources. Existing research including federated adopts a centralized architecture for where server aggregates updates clients/workers. The drawbacks such as performance bottleneck, poor scalability single point of failure. In this paper, we propose novel decentralized approach, E-Tree, which makes use well-designed tree structure imposed locations orders aggregation optimally designed improve training convergency accuracy. particular, design an efficient device clustering algorithm, named by KMA, E-Tree taking into account distribution devices well distance. Evaluation results show significantly outperforms benchmark approaches Gossip under NonIID in terms accuracy convergency.