Search papers, labs, and topics across Lattice.
This study explores the potential of physical noise in quantum hardware as a native regularizer for photonic hybrid quantum-classical neural networks (PHQCNNs), challenging the conventional view of noise as merely a nuisance. By employing a genetic algorithm to optimize noise parameters during training on datasets like Iris, Digits, and MNIST, the authors demonstrate that noise can enhance validation accuracy for certain datasets while degrading performance on others. The findings reveal that physical noise can serve as a beneficial regularizer, though its effectiveness is highly dataset-dependent, suggesting a nuanced approach to noise in quantum neural network training.
Physical noise can enhance the performance of quantum neural networks, but its benefits are not universal across all datasets.
Physical noise in near-term quantum hardware is usually treated as a nuisance to suppress. We ask whether it can instead act as a hardware-native regularizer for photonic hybrid quantum-classical neural networks (PHQCNNs), analogous to noise-injection regularization in classical deep learning. Using Quandela's Perceval simulator and the MerLin framework, we build PHQCNNs for Iris, Digits, and MNIST and inject Perceval's seven-parameter physical noise model directly into training. A genetic algorithm searches the six continuous noise dimensions and 1 boolean parameter to find, per dataset, the configuration maximizing validation accuracy, compared against a noiseless baseline across five seeds. GA-tuned noise yields modest accuracy gains on Iris (+0.82pp) and Digits (+1.45pp), but a clear degradation on MNIST (-1.21pp). Per-parameter sweeps show that no individual noise parameter is consistently beneficial, motivating the joint search, while a second-order loss expansion shows that physical noise induces a Tikhonov-like regularization term whose effect is dataset-dependent. Physical photonic noise can thus act as a free regularizer, but not universally.
