Privacy-Aware Offloading Strategy via Self-Supervised Feature Mapping in the End-Edge-Cloud System

Abstract

In the Internet of Things (IoT) era, the pervasive application of tremendous end devices puts forth an unprecedented demand for data processing. To address this challenge, the end-edge-cloud system has emerged as a solution, where task offloading plays a crucial role in efficiently allocating computing resources. Meanwhile, driven by the growing social awareness of privacy, privacy-aware task offloading methods have attracted significant attention. However, existing privacy-aware task offloading methods face various limitations, such as being applicable to specific scenarios, poor transfer ability of offloading strategies, etc . This paper studies the privacy-aware task offloading problem in the end-edge-cloud system and proposes PATO , a P rivacy- A ware T ask O ffloading strategy. PATO consists of two core modules. Specifically, a novel self-supervised feature mapping module transforms sensitive information via complex unidirectional mapping. Subsequently, a DRL-based decision-making module is trained to utilize transformed information to make task offloading decisions. Subtly combining the self-supervised feature mapping module and the DRL-based decision-making module, the proposed PATO addresses both privacy protection and task offloading challenges. Furthermore, PATO is designed as a general solution for task offloading problems and exhibits good transfer ability.