Quantum-inspired annealing-based semantic-aware optimization for sustainable cloud-edge intelligence

The swift development of artificial intelligence (AI) models within cloud-edge systems has intensified energy usage and created sustainability challenges, prompting concerns about sustainable AI. Recent studies have investigated quantum-inspired optimization and resource orchestration to improve efficiency; however, the majority of current approaches regard data samples and model updates as equally significant, overlooking their semantic impact on learning results. This research presents Semantic–Energy Aware Quantum-Inspired Adaptive Learning (SEA-QAL), an optimization methodology addressing semantic relevance, energy consumption, and predictive efficacy in distributed cloud–edge AI frameworks. In contrast to current energy-aware optimization techniques, SEA-QAL directly integrates semantic utility assessment with quadratic interaction-aware decision optimization via a closed-loop sustainability controller. The approach includes a semantic evaluation system that assesses update impact through gradient sensitivity, prediction uncertainty, and attention-driven representational signals. Semantic utility functions as an indicator of learning significance rather than a comprehensive measure of semantic meaning. These indicators are employed in energy cost modelling to develop a loss-energy-semantic Quadratic Unconstrained Binary Optimization (QUBO) framework. The resultant choice problem is addressed with a quantum-inspired annealing method that combines probabilistic state transitions with tunneling-like exploration in a classical computing framework. The approach improves optimization by including semantic utility, reducing unnecessary computation and communication costs while maintaining the model’s generalization capabilities. Comprehensive experiments conducted in cloud-edge workload contexts, including perception-oriented, sensor-based, and IoT-edge evaluations, demonstrate improvements in sustainability indicators, such as reduced energy usage and communication overhead, while achieving comparable or enhanced prediction accuracy relative to benchmarks. Furthermore, the evaluation of scalability with augmented edge-node configurations demonstrates the effectiveness of SEA-QAL for medium-scale cloud-edge deployments.