Bayesian Phase Estimation via Active Learning

Yuxiang Qiu, Min Zhuang, Jiahao Huang, Chaohong LeearXiv 2021 · DOI · arXiv

Abstract

Bayesian estimation approaches, which are capable of combining the information of experimental data from different likelihood functions to achieve high precisions, have been widely used in phase estimation via introducing a controllable auxiliary phase. Here, we present a non-adaptive Bayesian phase estimation (BPE) algorithms with an ingenious update rule of the auxiliary phase designed via active learning. Unlike adaptive BPE algorithms, the auxiliary phase in our algorithm is determined by a pre-established update rule with simple statistical analysis of a small batch of data, instead of complex calculations in every update trails. As the number of measurements for a same amount of Bayesian updates is significantly reduced via active learning, our algorithm can work as efficient as adaptive ones and shares the advantages (such as wide dynamic range and perfect noise robustness) of non-adaptive ones. Our algorithm is of promising applications in various practical quantum sensors such as atomic clocks and quantum magnetometers.

Introduction

Quantum phase estimation (QPE) serves as the core mechanism for ultra-precise sensors, including atomic clocks and quantum magnetometers. To estimate an unknown phase $\phi$, researchers frequently employ Bayesian Phase Estimation (BPE), a statistical framework that iteratively updates the probability distribution of the phase as experimental data is acquired. Traditionally, BPE implementations fall into two categories: adaptive and non-adaptive.

Adaptive (online) BPE optimizes the next measurement step based on all previous outcomes. While highly efficient in terms of measurement count, it demands significant real-time computational resources between every measurement, which can be a bottleneck in high-speed experimental settings. Conversely, non-adaptive (offline) BPE utilizes a pre-determined sequence of auxiliary phases. This approach is simpler to implement and exhibits superior robustness against noise, but it is inherently inefficient, often performing "uninformative" measurements that do not significantly refine the phase estimate.

Methodology: BPE via Active Learning (BPE-AL)

To address these limitations, we propose Bayesian Phase Estimation via Active Learning (BPE-AL). This method retains the structural simplicity of non-adaptive BPE while achieving the efficiency of adaptive techniques through a "pre-learning" stage.

Schematic comparison between standard BPE (a) and the proposed BPE-AL (b) architectures. The BPE-AL framework incorporates an Active Learning block to filter measurement steps.

The BPE-AL process begins with a pre-scan: a small batch of measurements ($M$ repeats) is performed across a range of auxiliary phases. The algorithm then applies entropy-based sampling to these results. By calculating the standard deviation ($\sigma_n$) or entropy of the outcomes for each phase, the system identifies "informative" indices—those where the outcome is most uncertain and thus provides the most information about the unknown phase.

In the subsequent estimation phase, physical measurements are only conducted for these selected informative indices ($n \in \{\tilde{n}\}$). For uninformative steps, the algorithm utilizes predicted outcomes derived from the learning phase. These filtered data points are then processed through a standard Bayesian update rule to narrow the probability distribution of the phase.

Results and Performance

Numerical simulations were conducted using a two-level Ramsey/Mach-Zehnder interferometer model to evaluate BPE-AL. The results demonstrate that BPE-AL can reduce measurement time by up to 85% compared to conventional BPE without sacrificing precision.

Performance metrics of BPE-AL. (c) Mean errors and (d) uncertainties reach the Standard Quantum Limit (SQL). (f) Comparison of measurement numbers shows the significant efficiency gain of BPE-AL over standard BPE.

As shown in the performance plots, the mean errors and uncertainties of BPE-AL scale identically to standard BPE, reaching the Ghosh bound at the SQL. Crucially, the algorithm maintains a full $2\pi$ dynamic range, avoiding the failures near phase boundaries often seen in some adaptive methods. Furthermore, BPE-AL demonstrates remarkable robustness; in the presence of simulated depolarization and phase noise, the algorithm's performance remains stable. This robustness is adjustable: as noise levels increase, the number of informative phases ($k'$) can be slightly increased to maintain precision.

Conclusion

BPE-AL represents a significant advancement in quantum sensing methodology. By leveraging active learning to identify and skip redundant measurements, it provides a high-efficiency path to phase estimation that is both computationally light and physically robust. This "plug-and-play" module is compatible with various quantum hardware platforms, such as cold atom ensembles or NV centers, making it a promising tool for the next generation of practical quantum sensors.

Summary updated

Abstract

Bayesian estimation approaches, which are capable of combining the information of experimental data from different likelihood functions to achieve high precisions, have been widely used in phase estimation via introducing a controllable auxiliary phase. Here, we present a non-adaptive Bayesian phase estimation (BPE) algorithms with an ingenious update rule of the auxiliary phase designed via active learning. Unlike adaptive BPE algorithms, the auxiliary phase in our algorithm is determined by a pre-established update rule with simple statistical analysis of a small batch of data, instead of complex calculations in every update trails. As the number of measurements for a same amount of Bayesian updates is significantly reduced via active learning, our algorithm can work as efficient as adaptive ones and shares the advantages (such as wide dynamic range and perfect noise robustness) of non-adaptive ones. Our algorithm is of promising applications in various practical quantum sensors such as atomic clocks and quantum magnetometers.
BibTeX
@misc{qiu2021bayesian,
  author = {Yuxiang Qiu and Min Zhuang and Jiahao Huang and Chaohong Lee},
  title = {{Bayesian Phase Estimation via Active Learning}},
  year = {2021},
  eprint = {2107.00196},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2107.00196},
}
Figure