Abstract
Classical IoT scheduling methods, such as Greedy heuristics, Genetic Algorithms, Particle Swarm Optimisation, and Reinforcement Learning have difficulty in finding an optimistic compromise between latency and energy usage since they stay local search, slow convergence, and cannot express the hidden task resource interaction found in heterogeneous and dynamic Internet of Things settings. These constraints result in inefficient performance when workloads change, when workloads vary in round-trip time (RTT), and under mass distributed deployments. To address them, the quantum-inspired latent scheduling framework presented in this paper combines learning a latent representation with tunnelling-based global Optimisation. The latent encoder unravels small-task resource embeddings that uncover latent interactions, whereas the quantum-inspired search leverages superposition-driven exploration to circumvent local minima and adapt to changing network states. Although the dataset is benchmark-based, it captures realistic RTT variability patterns reported in large-scale IoT systems. Experimental results show that the proposed model reduces latency by up to 25% and energy consumption by up to 28% compared to classical baselines. The framework has been tested using convergence analysis and scalability tests to support its superiority in multi-objective Optimisation. These results can guide the future of quantum-inspired latent modelling to be a scalable, interpretable and energy-efficient next-generation IoT scheduling technique.