Balancing Privacy and Performance: Machine Learning Security in Distributed Edge Networks

Abstract

Edge computing has transformed how machine learning models are deployed across distributed networks. As computing power gets closer to where data is generated, protecting privacy is becoming more and more important. This review looks at how edge computing and privacy-preserving machine learning come together, especially the trade-offs between privacy, performance, and usability. Past studies have been organized into three main areas: algorithmic approaches like federated learning and differential privacy, cryptographic methods such as secure multi-party computation and homomorphic encryption, and hardware-based techniques that rely on trusted execution environments. This analysis reveals a persistent deployment gap between mathematically elegant privacy guarantees and the practical constraints of resource-limited edge devices, where battery life, latency, and computational overhead force uncomfortable compromises. Key challenges have been identified in multi-objective optimization and evaluate existing balancing mechanisms and propose a conceptual framework that interprets privacy-performance trade-offs through multi-dimensional optimization, clarifying when certain techniques work better depending on deployment scenarios. This review brings together what has been achieved so far and highlights where the field is heading - toward building scalable, regulation-compliant, and efficient privacy-preserving AI systems that can work smoothly in distributed edge environments.