Data Management Strategies for Deep Learning-Based Quantum Computing Simulation

Abstract: Large-scale quantum simulation faces exponential growth in data volume due to the (2n)-dimensional Hilbert space, imposing severe storage, bandwidth, and data management constraints on classical computing systems. While deep learning offers a promising route for approximating quantum states and accelerating simulations, its performance is highly sensitive to data representation, sampling, and storage strategies. Here, we present a data-centric framework for classical deep learning-based quantum simulation, emphasizing hierarchical representations, adaptive sampling, noise-aware training, and metadata-driven storage. Our approach enables physically constrained, sample-efficient, and robust learning while minimizing storage overhead. Simulation studies in both quantum and radiology-inspired decision support contexts demonstrate that structured data management reduces memory requirements by orders of magnitude, improves predictive accuracy, enhances robustness to noise, and facilitates integration of hybrid datasets. These results highlight the critical role of principled data management in enabling scalable, reliable learning-accelerated scientific simulation systems.