January 05, 2024 | Jonathan Z. Lu,1,2,∗ Lucy Jiao,1 Kristina Wolinski,1,† Milan Kornjača,2 Hong-Ye Hu,1 Sergio Cantu,2 Fangli Liu,2 Susanne F. Yelin,1 and Sheng-Tao Wang2
The paper proposes a hybrid digital-analog learning algorithm on Rydberg atom arrays, combining the practical utility and near-term realizability of quantum learning with the rapidly scaling architectures of neutral atoms. The algorithm requires single-qubit operations in the digital setting and global driving according to the Rydberg Hamiltonian in the analog setting. The authors perform a comprehensive numerical study of the algorithm on both classical and quantum data, using handwritten digit classification and unsupervised quantum phase boundary learning as representative problems. They show that digital-analog learning is feasible in the near term, requires shorter circuit depths, and is more robust to realistic error models compared to digital learning schemes. The results suggest that digital-analog learning could open a promising path towards improved variational quantum learning experiments in the near term. The paper also discusses the physical hyperparameters of the digital-analog model and provides evidence of its benefits over purely digital learning schemes, particularly in terms of noise robustness and the ability to learn complex tasks requiring more expressivity.The paper proposes a hybrid digital-analog learning algorithm on Rydberg atom arrays, combining the practical utility and near-term realizability of quantum learning with the rapidly scaling architectures of neutral atoms. The algorithm requires single-qubit operations in the digital setting and global driving according to the Rydberg Hamiltonian in the analog setting. The authors perform a comprehensive numerical study of the algorithm on both classical and quantum data, using handwritten digit classification and unsupervised quantum phase boundary learning as representative problems. They show that digital-analog learning is feasible in the near term, requires shorter circuit depths, and is more robust to realistic error models compared to digital learning schemes. The results suggest that digital-analog learning could open a promising path towards improved variational quantum learning experiments in the near term. The paper also discusses the physical hyperparameters of the digital-analog model and provides evidence of its benefits over purely digital learning schemes, particularly in terms of noise robustness and the ability to learn complex tasks requiring more expressivity.