Blockchain-escorted distributed deep learning with collaborative model aggregation towards 6G networks

Abstract

The 6th generation (6G) network targets the Internet of Everything (IoE) implementation, and Distributed Deep Learning (DDL) can promote this progress with innovative performance in generating intelligence. Meanwhile, the 6G networks support ultra-reliable and low-latency communication (uRLLC) and thus can further elevate the DDL performance to empower the IoE development. However, DDL designs mostly focus on individual areas and yield separate intelligence which is insufficient for the IoE; besides, DDL platforms are usually managed in the centralized fashion, which is vulnerable for data preservation and task execution; the complexity of 6G networks involving heterogeneous devices and relations aggravates issues about reliability and efficiency of DDL. To this end, we propose a novel BC-escorted 6G-based DDL design for trustworthy model training. In this system, the 6G network design is utilized for system-wide uRLLC; non-homogeneous edge devices are grouped up with weighted consideration for DDL to train CNN models; macro base stations (MBSs) and small base stations (SBSs) jointly provide two-tiers parameter aggregation to elevate the knowledge level; a dual-driven BC consensus is designed to verify tasks and models; users anyplace can retrieve models via the BC nodes for object detection. The proposed design is evaluated in comparison with Cloud-based and P2P-based DDLs, and the results demonstrate better performance on accuracy and latency achieved in the proposed system.