QuACK: Accelerating Gradient-Based Quantum Optimization with Koopman Operator Learning

Quantum optimization, a key application of quantum computing, has traditionally been stymied by the linearly increasing complexity of gradient calculations with an increasing number of parameters. This work bridges the gap between Koopman operator theory, which has found utility in applications because it allows for a linear representation of nonlinear dynamical systems, and natural gradient methods in quantum optimization, leading to a significant acceleration of gradient-based quantum optimization. We present Quantum-circuit Alternating Controlled Koopman learning (QuACK), a novel framework that leverages an alternating algorithm for efficient prediction of gradient dynamics on quantum computers. We demonstrate QuACK's remarkable ability to accelerate gradient-based optimization across a range of applications in quantum optimization and machine learning. In fact, our empirical studies, spanning quantum chemistry, quantum condensed matter, quantum machine learning, and noisy environments, have shown accelerations of more than 200x speedup in the overparameterized regime, 10x speedup in the smooth regime, and 3x speedup in the non-smooth regime. With QuACK, we offer a robust advancement that harnesses the advantage of gradient-based quantum optimization for practical benefits.

Paper

Similar papers

Peer review

Reviewer ePBi7/10 · confidence 4/52023-06-21

Summary

This paper utilizes Koopman operator theory to accelerate the optimization of quantum circuits. Stability analysis is performed and the benefit of alternating between standard and Koopman optimization is discussed. A number of experiments are performed, across several different areas of interest to the quantum computing community, with Koopman optimization performing significantly faster than standard approaches. The effect of noise is studied, with QuACK again finding a benefit.

Strengths

1. This paper motivates the need for efficient, gradient based methods for optimizing quantum circuits. 2. This paper shows, via a number of experiments, the potential of Koopman operator theory to accelerate training, over a number of different regimes. 3. This paper tests multiple implementations for computing the Koopman spectra, including DMD, sliding DMD, and machine learning implementations. 4. The Appendix contains many details on quantum computing and the architectures used (the availability of this information should be mentioned in the main text).

Weaknesses

1. While Theorem 5.1 is true, in the sense that any finite numerical implementation of Koopman mode decomposition will find eigenvalues that are not exactly equal to 1, that does not mean that the eigenvalues computed by DMD will not be very close to 1 (see Redman et al., 2021 Fig. 4 for example of eigenvalue being 1.000), and that the associated Koopman mode could be a good approximation to a true fixed point. 2. Similarly, Theorem 5.2 and Corollary 5.4, while true, does not imply that QuACK will not lead the quantum circuit to converge to a different local minima than the gradient-based optimizer. This should be noted explicitly. 3. Fig. 3 shows that the Koopman based methods converge very quickly to the asymptotic energy and test accuracy - from just the plots alone it seems like the speed-up should be greater than the reported 4x. Why is this the case? 4. The Koopman learning section should be expanded to include mention of the Koopman mode decomposition (Mezic, 2005). Additionally, the need for data to computed the Koopman mode decomposition should be motivated in Sec. 3, before the QuACK algorithm (which needs data from the gradient based optimizer) is presented. The interested reader should be referred to more detailed discussion in Budisic et al., 2012 and Brunton et al., 2022. Finally, it should be noted that the idea of time-delays is related to Taken's theorem, with the implementation presented in Eq. 1 being Hankel DMD (Arbabi and Mezic, 2017). The sliding window DMD method is similar to streaming DMD (Hermati et al., 2014; Giannakis et al., 2023). 5. While this is, to my knowledge, the first approach of using Koopman operator theory to the study of quantum machine learning, it would be appropriate to cite more of the work on studying machine learning using Koopman operator theory (e.g., Mohr et al., 2020; Naiman and Azencot, 2021; Šimánek et al., 2022; Liang et al. 2022). Additionally, the previous work that has optimized neural networks using Koopman operator theory (Dogra and Redman, 2020; Tano et al., 2020) should be discussed in more detail (especially when presenting the speed-up $s$, which was defined similarly by Dogra and Redman, 2020, and when discussing the alternating strategy of Koopman training, which Tano et al. 2020 also utilized). Work using Koopman to study iterative algorithms (Dietrich et al., 2020; Redman et al., 2022) more generally should also be cited when mentioning that "the optimization trajectory...is a dynamical system" (line 52). Relatedly, the references with respect to Koopman operator theory need to be updated. Extended-DMD (EDMD) refers to a specific algorithm, developed by Williams et al. 2015. This was not referenced when EDMD was mentioned in the paper (line 80), and all other papers cited described extensions to DMD that are not EDMD. The connection between DMD and Koopman mode decomposition should be noted (Rowley et al., 2009). 6. Is it true that classical backpropagation is independent of $n_{params}$? I think you mean that you don't need to do individual perturbations along the dimension of each parameter, but this needs to be clarified. MINOR COMMENTS: 1. The speedup ratio of Fig. 2b and c seem to increase with the number of parameters. Is this expected? 2. Describing Koopman as "renowned for accelerating/predicting nonlinear dynamical systems" (lines 4 and 54) is a little overgeneral. Perhaps, saying "which has found utility in applications, as it allows for a linear representation of nonlinear dynamical systems," would be more specific and accurate. 3. It is mentioned that the quantum natural gradient is linked to imaginary time in the Introduction, but this is not discussed in overview section of Quantum Natural Gradient in Sec. 3. 4. There are a few minor typos: i. The question "How can we accelerate...", which is not a yes/no question, is posed (line 48), to which you answer "in the affirmative". ii. You list 3 speedups, but four conditions (lines 61, 62). iii. "of the system is called wave function" (line 88) iv. "the a layerwise partitioning" (line 299). Does this refer to the layer Koopman partitioning of Dogra and Redman, 2020?

Questions

1. Why does the 4x speedup reported for Fig. 3 not match the qualitative increase in speed-up from the figure? 2. Is it true that classical backpropagation is independent of $n_{params}$? 3. The speedup ratio of Fig. 2b and c seem to increase with the number of parameters. Is this expected? 4. Does "the a layerwise partitioning" (line 299) refer to the layer Koopman partitioning of Dogra and Redman, 2020?

Rating

7: Accept: Technically solid paper, with high impact on at least one sub-area, or moderate-to-high impact on more than one areas, with good-to-excellent evaluation, resources, reproducibility, and no unaddressed ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

The authors do a sufficient job addressing their limitations in the Limitations subsection of Sec. 4. They also study several different examples, across multiple different subregions of quantum computing, as well as different noise condition, providing strong evidence for their claims that Koopman training accelerates quantum optimization.

Reviewer sYsu6/10 · confidence 3/52023-06-22

Summary

This paper proposes QuACK, which leverages Koopman operator learning theory to accelerate gradient-based quantum optimization and machine learning. QuACK utilizes the sliding window DMD to enrich the embedding, and neural networks to learn and predict the nonlinear dynamics. Experiments on quantum Ising model, quantum chemistry and quantum machine learning show QuACK’s remarkable speedup and robustness. Also, theoretical analysis of its stability, convergence and complexity is provided.

Strengths

1. The proposed QuACK can significantly accelerate gradient-based quantum optimization by predicting the gradient dynamics, which may help deploy gradient-based quantum algorithms on NISQ devices. 2. Extensive experiments on quantum Ising model, quantum chemistry and quantum machine learning are conducted. Besides, speedup is verified for quantum natural gradients, in overparameterized scenes, and on IBM quantum devices. 3. Theoretical analysis of QuACK’s stability, convergence, complexity and speedup is provided.

Weaknesses

1. $\Theta(t_{d+1})\approx K\Phi(\Theta(t_0))$ in page 5 seems to lack theoretical support, since the koopman operator is $Kg(x(t))=g(x(t+1))$, which is different from the given formula. 2. Though gradient-based optimization with quantum computers can be costly and time consuming, an extra experiment on NISQ devices for gradient-based optimization is necessary for validating the feasibility of QuACK on realistic scenes.

Questions

1. The parameter $n_{sim}$ seems not well-defined in the paper. What is the specific meaning of it in Algorithm 1? 2. Have you compared QuACK with gradient-free methods like COBYLA?

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

Yes.

Reviewer kkUu6/10 · confidence 5/52023-07-03

Summary

The paper presents an innovative framework known as Quantum-circuit Alternating Controlled Koopman learning (QuACK), which effectively combines Koopman operator theory and natural gradient methods in quantum optimization. Through the utilization of an alternating algorithm, QuACK allows for the efficient prediction of gradient dynamics on quantum computers, resulting in a notable acceleration of variational quantum algorithms (VQAs). The authors provide empirical evidence showcasing the effectiveness of QuACK across diverse VQA tasks, including those involving noisy environments.

Strengths

The paper's strength lies in its contribution to accelerating the optimization of variational quantum algorithms (VQAs), which is crucial for leveraging the capabilities of modern quantum machines in practical problem-solving. The proposed QuACK addresses this challenge and presents intriguing findings. By incorporating Koopman operator theory and deep neural networks, QuACK offers noticeable speed improvements compared to conventional optimization methods like the parameter shift rule. The numerical results provided in the paper demonstrate the efficacy of QuACK.

Weaknesses

The primary concern of the paper lies in the stability and robustness of QuACK. Algorithm 1 suggests that the performance of QuACK may heavily rely on the hyperparameters n_sim and n_DMD. Determining the appropriate values for these hyperparameters is an empirical process and may require additional hyperparameter optimization, introducing a potential computational burden. While the authors have conducted extensive numerical analysis, it remains unclear whether an improper setting of these hyperparameters could undermine the advancements of QuACK in optimizing modest-sized VQAs that aim to achieve quantum advantages, particularly in scenarios involving 100+ qubits. Furthermore, the issue of stability is exacerbated by QuACK's sensitivity to the system and shot noise. As demonstrated in the study by Du et al. (PRX Quantum 2.4, 2021: 040337), gradient-based methods can exhibit divergence under significant noise and low shot numbers. It remains uncertain whether QuACK demonstrates greater robustness compared to standard optimization strategies in such challenging scenarios. Addressing these stability and robustness concerns would augment the applicability and reliability of QuACK in practical quantum optimization tasks. A more in-depth discussion of these issues would be preferable. In addition, there are some minor issues with the paper: - The font size in Figure 1 is too small, making it challenging to read. - The relation between n_sim and m is not clearly defined in the paper, and clarification is needed. - Corollary 5.4 could be removed since it is redundant given that the optimization of VQAs is generally non-convex. - The format of references does not adhere to the standard style in NeurIPS. - Several references are missing, including: 1. You, Xuchen, et al. arXiv:2303.14844. 2. García-Martín, D., Larocca, M., & Cerezo, M. arXiv:2302.05059. 3. Wang, Xinbiao, et al. arXiv:2208.14057.

Questions

The concerns listed in *Weakness* should be adequately addressed. Satisfactory responses to these points will further enhance the quality of the paper.

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

5: You are absolutely certain about your assessment. You are very familiar with the related work and checked the math/other details carefully.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

The authors partially addressed the limitations of QuACK.

Reviewer M6jg7/10 · confidence 3/52023-07-05

Summary

The paper presents an algorithm that effectively "extends" $k$ gradients updates to $k'$ gradient updates by fitting a neural koopman operator. The additional cost is made worth it because the algorithm is derived in a quantum machine learning setting where quantum gradients are expensive and limited and classical operations are comparatively free and unlimited. The algorithm successfully speeds up Quantum Natural Gradient, in particular in the quantum overparametrzation regime (200x) but even in nonsmooth and noisy regimed it achieves 2x to 3x speedup. Convergence proofs for convex problems and some other theoretical results augment the derivation and empirical evaluation.

Strengths

Originality: people don't use koopman operators enough, using them as a framework to perform a $m$-step extension of $k$ expensive gradient steps is very clever and I might steal the idea for some of my own work post review period Quality: Argument is clear and sufficient analysis is done to sanity check and gain insight into the algorithm, but I would call this an empirical paper still, not in the least due to the training tricks required to make the neural koopman estimate work. For the empiricism, full marks, good ablations in the appendix, error bars where it matters. Bonus points for an attempt on real quantum experiments, even if the experiments are kind of tangential to the method presented (gradient free vs. gradient dynamics estimation) Clarity: well written, easy to understand Significance: speaking as an ML person, I think this is a significant contribution to QML, in particular if we remain constraint by quantum operations

Weaknesses

- what is the wallclock time of each method? this is a relatively crucial detail I am missing for context

Questions

- what is the distinction between previous "neural koopman" works (e.g. lusch et al) and this work? is it "just" that it is applied to QML learning dynamics?

Rating

7: Accept: Technically solid paper, with high impact on at least one sub-area, or moderate-to-high impact on more than one areas, with good-to-excellent evaluation, resources, reproducibility, and no unaddressed ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

4 excellent

Presentation

3 good

Contribution

3 good

Limitations

I think limitations are well stated

Reviewer ePBi2023-08-11

Response to reviewers

Thank you for your detailed rebuttal and for including items discussed in the review into the paper. I believe the paper has been strengthened and I feel that my questions/concerns have been sufficiently clarified. Based on this and reading all reviews from other reviewers, as well as the authors' responses to these reviewers, I feel confident in the quality of this paper. I will increase my score.

Authorsrebuttal2023-08-14

Response to Reviewer ePBi

We appreciate Reviewer ePBi for reading all the reviews and our responses. We would also like to thank the Reviewer for acknowledging the quality of our paper and increasing our score. If there is any follow-up comment or suggestion, please let us know and we would be glad to discuss and elaborate more.

Reviewer sYsu2023-08-17

Thanks for addressing most of my concerns. Based on the other reviewers' opinions and your rebuttals, I will raise my score.

Reviewer kkUu2023-08-18

I thank the authors for their careful response, which adequately addresses all my concerns.

Program Chairsdecision2023-09-21

Decision

Accept (spotlight)

© 2026 NYSGPT2525 LLC