Efficient Reconstruction of Wigner Function from Marginals with Deep Generative Model
Po-Hsuan Chen, Li-An Lo, Tzu-Chia Liu +2 more
The Wigner function is one of the most important tools for characterizing quantum states in continuous-variable and quantum optical systems, with its negativity serving as a strong indicator of nonclassicality. However, its experimental reconstruction from measurable raw data is a tedious task requiring substantial experimental resources. On the other hand, machine learning has made immense strides and been successfully applied across various scientific disciplines. Building on our previous development of a ResNet-based deep generative model (DGM), here we further optimize the DGM to reconstruct complex Wigner functions by processing merely three experimentally measurable marginal distributions. We validate our approach using two sophisticated cavity quantum electrodynamics models, where coherent light-matter interactions introduce intricate phase-space features. Our results reveal that the DGM is capable of capturing statistical features and physical characteristics within the marginals and reproducing the intricate patterns of the Wigner functions with high accuracy. Furthermore, we demonstrate its superior performance over a widely adopted denoising diffusion model, as well as its generalizability to the time evolution of states modestly beyond the training range. Consequently, our approach is expected to substantially reduce the experimental resources required for quantum state tomography.
IEEE Transactions on Quantum Engineering•Jan 1, 2026•0 citations