Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
Q: Fachverband Quantenoptik und Photonik
Q 21: Quantum Computing and Simulation III
Q 21.7: Vortrag
Dienstag, 3. März 2026, 12:45–13:00, P 10
Boosting Classification with Quantum-Inspired Augmentations — Matthias Tschöpe1, Vitor Fortes Rey1,2, Sogo Pierre Sanon1, Paul Lukowicz1,2, Nikolaos Palaiodimopoulos1,2, and •Maximilian Kiefer-Emmanouilidis1,2 — 1DFKI Kaiserslautern — 2RPTU Kaiserslautern-Landau
Small quantum gate perturbations, common in quantum hardware but absent in classical computing, are typically viewed as errors, yet they may serve as a form of data augmentation and offer advantages in quantum machine learning. In this work, we study random Bloch sphere rotations, fundamental SU(2) transformations, as a simple quantum-inspired augmentation method for classical image classification. Unlike standard techniques such as flipping or cropping, these transformations lack intuitive spatial interpretation. Rather than using quantum models or quanvolutional layers, we apply small-angle Bloch rotations directly to classical data and evaluate their effect. Experiments on the ImageNet dataset show consistent performance gains, including a 3% improvement in Top-1 accuracy, a 2.5% gain in Top-5 accuracy, and an increase in F1 score from 8% to 12% over standard augmentation pipelines. We also explore stronger unitary transformations, which produce visually unrecognizable images with potential relevance to privacy. However, we find no measurable improvements in differential privacy and discuss the implications.
Keywords: Quantum Machine Learning; Data Augmentation; Image Classification; Quantum-Inspired; Quantum Computing