Year: 2025
Type: article
Abstract: The convergence of edge and cloud intelligence necessitates adaptive, explainable, and resource-efficient decision-making frameworks capable of operating under dynamic and heterogeneous environments. Traditional deep-learning-driven edge intelligence models, while effective in perception and prediction, suffer from limited interpretability, high retraining overhead, and poor adaptability to fluctuating network and resource conditions. To address these challenges, this paper proposes an Adaptive Neuro-Symbolic Fusion Framework (ANSFF) for self-optimizing edge–cloud intelligence. The framework integrates lightweight neural inference at the edge with symbolic reasoning and policy optimization in the cloud, enabling both data-driven learning and logic-based decision validation. A reinforcement-based adaptive controller dynamically orchestrates computation across edge and cloud layers based on latency, energy constraints, workload intensity, and task criticality. Symbolic rules and domain knowledge are embedded into the learning process through logic-regularized optimization, enhancing explainability and stability. Extensive evaluations on distributed IoT and edge–cloud benchmark scenarios demonstrate that the proposed framework achieves higher decision accuracy, reduced inference latency, and lower energy consumption compared to conventional deep-learning-only and static edge–cloud architectures. Furthermore, neuro-symbolic integration significantly improves decision transparency and reduces retraining overhead, making the framework suitable for scalable, trustworthy, and autonomous edge–cloud intelligence systems.
Source: CompSci & AI Advances
Institution: Dambi Dollo University
Language: English
Topic: IoT and Edge/Fog Computing
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7204242433