Resource-Aware and Energy-Efficient Machine Learning for Distributed Edge-Cloud Systems: A Comprehensive Survey

Abstract

– The vast proliferation of Internet of Things (IoT) devices with constrained resources led to the adoption of distributed edge – cloud computing systems. Machine learning (ML) has become an important source for deploying intelligent services across many environments. But deploying and training ML models in resource-constrained edge devices presents significant challenges, including limited computational capability, restricted memory, energy constraints, and communication overhead. In this work a systematic survey of latest research in the design and deployment of resource-aware and energy-efficient machine learning models in distributed edge – cloud systems are presented. The survey categorizes existing approaches into model optimization methods, including pruning, quantization, knowledge distillation, and lightweight architecture design, as well as system-level strategies such as federated learning, task offloading, and collaborative inference. Furthermore, the paper analyses trade-offs among accuracy, latency, energy consumption, and communication cost, highlighting the impact of these techniques on real-world deployments. Recent advancements in adaptive and context-aware learning frameworks are also examined, emphasizing their role in dynamic resource management. In addition, open challenges such as heterogeneity of edge devices, scalability, privacy preservation, and robustness are critically discussed. Finally, the paper identifies limitations and future scope to lead future research directions aimed at enabling sustainable and scalable edge intelligence.