QuFrame: A Novel Encoding Ensemble Framework for Quantum Neural Networks
Tingting Li, Ziming Zhao, Liqiang Lu, Jianwei Yin · 2025
Quantum neural networks (QNNs) have gained significant attention in quantum machine learning. To apply QNNs to classical tasks on quantum computers, a key step is encoding classical data into quantum states, which directly affects the model’s abstraction ability and performance. However, existing encoding schemes face challenges related to encoding space, fault tolerance, and more. In this paper, we present QuFrame, a novel framework for encoding ensembles in QNNs that combines the strengths of various encoding methods. Specifically, we design a suite of pipelines to realize QuFrame, including 1) complex linear layer weighted encoding ensemble, 2) ℓ2normalization for matching the definition of the quantum state, and 3) weight optimization based on gradient descent. Moreover, we introduce a series of metrics to evaluate the quantum state distribution and formulate several quantum kernels to provide theoretical insights and visualizations for QuFrame. Experiments show that QuFrame improves the average accuracy and F1 score of QuFrame by 7.2%∼31.85% and 11%∼37.74%, compared with four typical quantum state encoding methods, respectively.