Deterministic Reflexes and Stochastic Planning for Decoupling Control and Learning in Resource Constrained Edge Offloading

Abstract

Current edge computing research highly favors complex deep learning models for resource management. However, these data-heavy approaches often conflict with the physical reality of edge devices, which demand low latency, low energy usage, and absolute predictability. This review critically examines the trade-offs between stochastic learning strategies and lightweight, deterministic alternatives. By analyzing recent literature, we highlight that simple, rule-based logic frequently outperforms complex learning algorithms in stability and response time. The findings suggest a need to shift design priorities away from unnecessary complexity and toward sufficiency, prioritizing reliability over raw intelligence at a high operational cost.