Optimizing Deep Neural Networks Using Quantum-Inspired Evolutionary Algorithms For High-Dimensional Data Classification

Abstract

High-dimensional data classification poses major challenges in terms of computational complexity, convergence stability, and generalization performance of traditional deep neural networks (DNNs). To address these challenges, this paper proposes a Quantum-Inspired Evolutionary Algorithm (QIEA)-based optimization framework for deep neural networks, enabling efficient training and superior accuracy in high-dimensional feature spaces. The proposed approach integrates quantum-inspired population dynamics—employing qubit representation, superposition, and probabilistic observation—to enhance exploration–exploitation balance in the optimization process. The QIEA adaptively adjusts DNN hyperparameters, including learning rate, weight initialization, and network depth, thereby minimizing overfitting and accelerating convergence. The framework is evaluated on multiple benchmark datasets such as CIFAR-10, MNIST, and UCI high-dimensional data, using performance metrics including accuracy, F1-score, and computational efficiency. Experimental results reveal that the QIEA-optimized DNN achieves a 4–8% improvement in classification accuracy, a 32% reduction in training time, and a 27% lower error rate compared to conventional gradient-based optimizers and metaheuristic algorithms like PSO and GA. Moreover, the model exhibits remarkable robustness against data noise and high feature dimensionality, proving its effectiveness for applications in bioinformatics, text analytics, and image recognition. The findings highlight the potential of quantum-inspired optimization to redefine deep learning efficiency in large-scale, data-intensive environments.