arXiv:2609.20523
Noise-Robust Quantum State Characterization for Remote State Preparation with Deep Learning
Bo Tang, Zixuan Liao, Hao Li, Yilin Yang, Jiani Lei, Zengya Li, Jing Qiu, Zhaohui Dong, Zhengyang Mao, Yuanhua Li, Yuanlin Zheng, Xianfeng Chen
Abstract
Quantum communication underpins secure information processing and scalable quantum networks. In particular, remote state preparation (RSP) enables efficient quantum state transfer, but accurately estimating target states under complex noise remains challenging. Here, we propose a Transformer-based Quantum State Characterizer (TQSC) model for noisy RSP experiments. Our model reconstructs experimentally prepared pure and mixed photonic polarization states from noisy measurements in complex scattering environments, while its attention patterns provide physically grounded insights into correlations among the measured observables. The method achieves a mean estimator-target fidelity exceeding 99.999% under complex scattering and dynamic Gaussian noise, while its robustness and generalization are further examined using Qiskit-simulated Bloch-ball states. Furthermore, in a practical MNIST image transmission task with held-out states, the decoded bit error rate is reduced from 50.34% to zero after TQSC post-processing. The TQSC model enables accurate tomographic characterization under dynamic noise and provides physically grounded post-hoc insights, holding promise for intelligent quantum information processing applications.
AI Summary
The paper studies supervised reconstruction of target single-qubit polarization states from noisy measurement records in photonic remote state preparation (RSP). A Transformer-based Quantum State Characterizer takes 13 inputs—coincidence counts, Alice/Bob singles counts in H/V/D/R settings, and acquisition time—and outputs a density matrix through a normalized Cholesky-factor parametrization. Training uses paired measurement records and theoretical target density matrices with a mean-squared-error loss. The experiment uses polarization-entangled telecom photons, multimode fiber (MMF) to introduce scattering, and phase modulation driven by Gaussian noise. The experimental state set comprises seven diagonal mixed states and one pure state. Without MMF, conventional tomography gives mean target fidelity $(99.25\pm0.20)\%$. For the pure-state noise test, MMF plus dynamic perturbations reduces fidelity to $(38.31\pm12.30)\%$. Across training and test data, the network reports mean estimator–target fidelity $0.99999497\pm8.78\times10^{-6}$; this measures agreement with target labels, not physical restoration of transmitted states. It slightly outperforms a parameter-matched MLP. Interpolation, leave-one-out, and Qiskit-simulated Bloch-ball tests yield fidelities above $99\%$. A held-out, 1176-bit MNIST transmission demonstration reports BER reduction from $50.34\%$ to zero through classical post-processing. Attention and token-ablation analyses examine dependence on measurement features. The stated limitations include fewer than 400 test samples, possible sensitivity to edge cases, the need for paired training examples, and unresolved scalability and robustness under more complex noise. Experimental validation is restricted to the tested state families and controlled noise conditions.
Central claim may not hold: The headline fidelity above 99.999% is reported for pooled training and test data across conditions, without a separate held-out-test mean establishing that performance under scattering plus dynamic noise. Agreement with intended theoretical target states validates supervised target inference, but does not independently validate reconstruction of the actual experimentally prepared noisy density matrices, substantially weakening the central tomography claim.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.