SECURING DATA PRIVACY IN EDGE CLOUD COLLABORATIVE SYSTEM

This paper explores the challenges and opportunities in edge-cloud collaborative systems for running Deep Learning tasks on resource-constrained IoT devices. It highlights the potential privacy issues introduced by third-party clouds in edge computing. The paper presents two main contributions: 1. It introduces a set of new attacks that allow an untrusted cloud to recover arbitrary inputs without accessing the edge device’s data, computations, or system permissions. 2. The paper empirically demonstrates that adding noise as a defense mechanism is ineffective against the proposed attacks. It then proposes two more effective defense methods. Overall, the study provides insights and guidelines for developing privacy-preserving collaborative systems and algorithms in the context of edge-cloud computing.