Communication-Efficient Multiuser AI Downloading via Reusable Knowledge Broadcasting

Abstract

<p>In-situ model downloading is a 6G service for real-time downloading of trained AI models to devices based on their tasks. The simultaneous downloading of diverse, high-dimensional models from an edge server to multiple devices over wireless links results in a communication bottleneck. To overcome the bottleneck, we propose the framework of model broadcasting and assembling (MBA), which represents the first attempt on leveraging reusable knowledge, referring to shared parameters among tasks/models, to enable parameter broadcasting to reduce communication overhead or latency. The MBA framework is designed to comprise two key components. The first, the MBA protocol, defines the system operations including parameter selection from an AI library, power control for broadcasting, and model assembling at devices. The protocol features the use of Shapely value as a metric for measuring parameters’ reusability. The second is the joint design of parameter selection and power control, which provides guarantees on devices’ model performance and aim to minimize the downloading latency. Experimental results on real-world datasets show that the MBA can substantially reduce the entire downloading latency as opposed to the traditional unicasting based model downloading.<br></p>