Quantum machine learning (QML) concerns the solution of machine learning problems via quantum computers. Recent years have seen an enormous interest in the field due to its high potential for overcoming limitations of classical ML for specific problems. Our research focuses on theoretical questions such as robustness and generalization in order to better understand and improve the capabilities of QML methods for supervised and reinforcement learning tasks. We develop regularization techniques to train quantum models that are inherently more robust against adversarial attacks and quantum noise, and that generalize better to unseen data. Key challenges arising in this context include the trade-off between efficiency and accuracy as well as the consideration of quantum data, which plays a crucial role in obtaining computational advantages.
Representative publications
- J. Berberich, T. Fellner, C. Holm - "The interplay of robustness and generalization in quantum machine learning" - Quantum Robustness in Artificial Intelligence, Springer book series Quantum Science and Technology, 2026. Link
- J. Berberich, D. Fink, D. Pranjic, C. Tutschku, C. Holm - "Training robust and generalizable quantum models" - Physical Review Research, 2024. Link
- N. Meyer, J. Berberich, C. Mutschler, D. D. Scherer - "Robustness and generalization in quantum reinforcement learning via Lipschitz regularization" - arXiv:2410.21117. Link
Collaborations
D. Dong (UT Sydney, AU)
C. Holm (Institute for Computational Physics, University of Stuttgart, DE)
C. Mutschler, D. D. Scherer (Fraunhofer IIS, Nürnberg, DE)
C. Tutschku (Fraunhofer IAO, Stuttgart, DE)