Bayesian Quantum Amplitude Estimation

Alexandra Ramôa 1,2,3, Luis Paulo Santos 1,2,3 · 2024

Read Bayesian Quantum Amplitude Estimation by Alexandra Ramôa 1,2,3, Luis Paulo Santos 1,2,3 (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-04394.txt.

Introduction

Alexandra Ramôa and Luis Paulo Santos published this paper in 2024. It improves quantum amplitude estimation, a core routine behind many quantum algorithms. The routine measures the strength of one part of a quantum state, and its accuracy sets the value of quantum speedups in finance and chemistry.

Their algorithm, BAE, takes a Bayesian route: it keeps a running belief about the answer and updates it with each measurement. On perfect hardware it reaches the best accuracy physics allows. On noisy hardware it learns the noise as it goes and adapts. An annealed variant adds robustness borrowed from statistics. Both versions split work across machines and can reuse earlier knowledge. In the authors' new benchmark, BAE beat rival methods on cost for both noisy and clean settings.

Catalog record

CREATORAlexandra Ramôa 1,2,3, Luis Paulo Santos 1,2,3
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE100 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-04394

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Bayesian Quantum Amplitude Estimation Preprint: APS/123-QED Alexandra Ramôa 1,2,3, Luis Paulo Santos 1,2,3 Abstract Quantum amplitude estimation is a fundamental routine that offers a quadratic speed-up over classical approaches. The original QAE protocol is based on phase estimation. The associated circuit depth and width, and the assumptions of fault tolerance, are unfavorable for near-term quantum technology. Subsequent approaches attempt to replace the original protocol with hybrid iterative quantum-classical strategies. In this work, we introduce BAE, a noise-aware Bayesian algorithm for QAE that combines quantum circuits with a statistical inference backbone. BAE can dynamically characterize device noise and adapt to it in real-time. Problem-specific insights and approximations are used to keep the problem tractable. We further propose an annealed variant of BAE, drawing on methods from statistical inference, to enhance statistical robustness. Our proposal is parallelizable in both quantum and classical components, offers tools for fast noise model assessment, and can leverage preexisting information. Additionally, it accommodates experimental limitations and preferred cost trade-offs. We show that BAE achieves Heisenberg-limited estimation and benchmark it against other approaches, demonstrating its competitive performance in both noisy and noiseless scenarios. I Introduction Quantum amplitude estimation (QAE) is a routine for estimating the measurement probability associated with a given substate for some wavefunction Brassard et al. 2002 . Notable applications are Monte Carlo integration Montanaro 2015 and estimation tasks Knill et al. 2007 , with applications in finance Woerner and Egger 2019 ; Gómez et al. 2022 ; Rebentrost et al. 2018 ; Miyamoto 2022 ; Stamatopoulos et al. 2020 ; Gómez et al. 2022 , chemistry Kassal et al. 2008 ; Katabarwa et al. 2021 ; Johnson et al. 2022 ) and machine learning Wiebe et al. 2016 ; Wiebe et al. 2015 ; Kerenidis et al. 2019 ; Wiedemann et al. 2022 . The original proposal for QAE achieves a quadratic quantum advantage by performing phase estimation on a quantum amplitude amplification (QAA) operator Brassard et al. 2002 . However, the complexity of the corresponding circuit is prohibitive for the currently available quantum devices, due to their susceptibility to noise and limited size. Alternatives have been proposed in the literature, often hybrid quantum-classical algorithms where simpler circuits are embedded in a classical feedback loop Aaronson and Rall 2020 ; Wie 2019 ; Suzuki et al. 2020 ; Wang et al. 2017 ; Nakaji 2020 ; Grinko et al. 2021 . The simpler circuits typically consist of a sequence of m non-controlled applications of the amplification operator, where m is an experimental degree of freedom controlled by the classical processing unit, possibly in an adaptive fashion. In this formulation, QAE is largely similar to common characterization tasks associated with superconducting qubits Ramôa 2024a . These tasks are related to precession dynamics, such as Larmor Malcom Levitt 2008 , Rabi Merlin 2021 and Ramsey Ramsey 1950 oscillations. Similarly, in photonic quantum computing, Mach-Zehnder interferometry gives rise to a similar framework Pezzé and Smerzi 2014 . All of these dynamics admit squared-sinusoidal descriptions, just like Grover circuits. Techniques applied to one of them likely transfer to the others, extending the applicability of QAE algorithms to tasks such as sensing or quantum gate implementation and calibration; and inversely, QAE can borrow from algorithms devised for those tasks. Importantly, unlike in many hybrid near-term quantum algorithms, such schemes can provably achieve the full quadratic advantage for QAE Aaronson and Rall 2020 ; Fukuzawa et al. 2023 . However, the implementation details remain an open question. They are determinant for performance factors such as the quantum cost offset, classical processing cost, parallelism, and noise resilience. While the aforementioned alternative algorithms require simpler circuits, most of them still assume ideal executions, considering at most sampling noise as a limitation. What is more, their performance is often assessed under that assumption, rendering their behavior unpredictable in the presence of noise. One framework capable of generalizing to the noisy case is Bayesian inference, which applies naturally to the iterative formulation of QAE. However, the involved processing costs are high. The circuit length is expected to increase exponentially with the iteration number Ferrie et al. 2012 ; Suzuki et al. 2020 , with higher precision requiring more iterations. In a naive strategy, the searching range and thus the optimization cost per iteration are expected to increase exponentially with the total iterations. On the other hand, the cost of a naive (deterministic) statistical approximation increases exponentially with the number of parameters. More efficient approximations are vital to ensure tractability; importantly, care must be taken lest they jeopardize scalability or correctness. In Wang et al. 2021 , a noise-aware approach to QAE based on Bayesian inference is proposed under a normality assumption. While this is a cost-effective approach for simple cases, it is not expected to generalize well to more complex ones, namely problems involving more elaborate noise models. In this work, we propose an approximate Bayesian algorithm that is capable of handling noise without making restrictive assumptions or incurring prohibitive costs. Problem tailored heuristics are used to cut down optimization costs while preserving the quantum advantage, bringing the optimization cost per iteration down to constant. By employing scalable and highly parallelizable statistical methods and other approximations, we further lower the classical processing costs without affecting the quantum advantage. The employed quantum circuits are of the simplest type presented in the literature; they do not require controlled or generalized versions of the Grover operator, nor any other qualitative complication as compared to Grover search. Our method offers quantifications of model merit with little extra cost, can incorporate available information such as past experimental data, and is flexible with respect to the cost trade-offs involved. For instance, one may trade in part of the quantum advantage for a lower classical cost, or for increased parallelism in the quantum sub-routine. To benchmark BAE, we numerically test its performance along with those of other QAE algorithms presented in the literature, and propose thorough and cost-effective methods to evaluate and analyze their performance. Additionally, inspired by a widely used method in statistics, we propose a variant of BAE where the quantum circuits are guided based on measures of statistical efficiency not directly related to the error or uncertainty. This further highlights the importance of this component, and offers a robust and systematic way to work with challenging models or data. We call this variant annealed Bayesian amplitude estimation, after the method it is inspired in, annealed likelihood estimation South et al. 2019 . The rest of the paper is organized as follows. Section II presents relevant background, including the original quantum algorithm, the classical counterpart, and alternative hybrid algorithms. In section III , BAE is presented. Section IV describes the numerical methods used to test the algorithms. The results of these tests are shown in section V . Finally, section VI discusses and summarizes the results, as well as directions for future research. II Background II.1 Amplitude estimation Quantum amplitude estimation (QAE) Brassard et al. 2002 is a process that estimates the parameter a in a wavefunction of the form \left\lvert\psi\right\rangle=\mathrm{\hat{A}}\left\lvert 0\right\rangle^{\otimes n}=\sqrt{a}\left\lvert\psi_{1}\right\rangle+\sqrt{1-a}\left\lvert\psi_{0}\right\rangle, (1) with a\in[0,1] and \left\lvert\psi_{1}\right\rangle , \left\lvert\psi_{0}\right\rangle projections of \left\lvert\psi\right\rangle into two different subspaces. \mathrm{\hat{A}} is a problem-dependent initialization operator. Despite being a probability, the parameter a

Read the complete text