Energy-Efficient Edge Learning via Joint Data Deepening-and-Prefetching

Abstract--The vision of pervasive artificial intelligence (AI) these services can be designed in a user-customized manner services can be realized by training an AI model on time usingby training an AI model using a target user's data before going real-time data collected by internet of things (IoT) devices. To this end, IoT devices require offloading their data to an edge out of date. A new paradigm of edge learning has emerged as a server in proximity. However, transmitting high-dimensional andviable solution such that an AI model can be quickly trained at voluminous data from energy-constrained IoT devices poses athe edge server collocated with a wireless access point instead significant challenge. To address this limitation, we proposeof a central cloud [2]. One key prerequisite of edge learning a novel offloading architecture, called joint data deepening-is that the data required for training should be offloaded to and-prefetching (JD2P), which is feature-by-feature offloadingthe edge server on time over a wireless channel, which can comprising two key techniques. The first one is data deepening, be a heavy burden to energy-constrained IoT devices [3]. Towhere each data sample's features are sequentially offloaded in the order of importance determined by the data embeddingcope with this issue, we propose a novel technique, called joint technique such as principle component analysis (PCA). By leveraging a data is terminated once the already transmitted features are sufficientembedding technique, each data sample can be represented for accurate data classification, resulting in a reduction in theby several features.