Comparative Study of Quantum and Classical Layers in Hybrid Quantum Neural Networks
Book Chapter
Etar, A, Tripathi, S, Soni, J et al. (2027). Comparative Study of Quantum and Classical Layers in Hybrid Quantum Neural Networks
. 685 LNICST 64-80. 10.1007/978-3-032-22542-9_5
Etar, A, Tripathi, S, Soni, J et al. (2027). Comparative Study of Quantum and Classical Layers in Hybrid Quantum Neural Networks
. 685 LNICST 64-80. 10.1007/978-3-032-22542-9_5
Hybrid quantum-classical neural networks (HQNNs) combine classical neural network components with parameterized quantum circuits to explore potential advantages of quantum computing for machine learning tasks. In this study, we investigate the performance of HQNNs on a regression problem from high–energy physics. We implement two model variants which consist of a hybrid classical-quantum architecture and a quantum-centric model, each paired with three different quantum circuit ansätze. Using a standardized preprocessing of the CERN dielectron dataset consisting of 100k events and 16 input features, we train and evaluate all six model-ansätz configurations over multiple runs. Our results indicate richer network expressivity and improved accuracy due to the classical harness over the quantum layer while purely quantum layers face challenges under NISQ constraints. In particular, models with sufficiently expressive ansätze can achieve performance comparable to classical baselines but training instability and limited entanglement in shallow circuits can hinder convergence. These findings highlight the importance of architecture design and initialization in HQNNs and suggest that current quantum layers may serve best as feature transformations within this hybrid pipeline. A workflow diagram and detailed schematic illustrations are provided to clarify the experimental setup.
