Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2610.32571

Classical Pre-Training of Large Shallow Parametrized Quantum Circuits through Pauli Propagation and deployment on Quantum Hardware

Saverio Monaco, Jamal Slim, Ran Xue, Dirk Krücker, Kerstin Borras

quant-phSubmitted Oct 23, 2026• Announced Oct 26, 2026• 0 citations
arXivHTMLPDF

Abstract

Scaling parametrized quantum circuits to a larger number of qubits is a central challenge in Quantum Machine Learning: gradient estimation via the parameter-shift rule quickly becomes prohibitively expensive, while gradient-free optimizers offer limited convergence guarantees. We apply the Pauli propagation framework to a broad class of Quantum Machine Learning problems in which the loss is expressed as a function of expectation values of the circuit's observables, while keeping the relevant quantity of interest, such as the quantum state or its samples, classically intractable to compute. By propagating the observables backward through the circuit, we obtain an explicit parametric expression for the loss that can be minimized entirely on classical hardware. We also present complementary strategies to make propagation tractable at large scale, including weight-based and degree-based truncations and pruning heuristics. We demonstrate the approach on two benchmarks run on 64-qubit circuits: Variational Quantum Eigensolver on a two-dimensional Ising model, and training of a discrete-type generative model for binary images. In both cases, the classically pre-trained parameters are deployed on the IBM quantum processor ibm_berlin. For generative learning, we additionally show that the circuits considered here require substantially fewer parametrized gates than IQP-style ansätze, reducing transpilation overhead and making deployment on current quantum hardware feasible.

AI Summary

The paper asks whether large, shallow parametrized quantum circuits can be trained classically using observable-based losses, then deployed on quantum hardware. It backward-propagates observables through the circuit to construct explicit trigonometric functions of the parameters, optimized with classical gradients. Pauli-weight and trigonometric-degree cutoffs control expression growth; two exact pruning rules remove branches that cannot contribute to the initial-state expectation value. Demonstrations use 64-qubit circuits on an $8\times8$ grid, with deployment on IBM’s `ibm_berlin`. For VQE on the two-dimensional transverse-field Ising model at $(J,h)=(1,2)$, depth-1 and depth-2 ansätze have 256 and 384 parameters. Training uses 1,000 Adam steps, with exact propagation at depth 1 and weight cutoff $w_{\mathrm{cut}}=12$ at depth 2. Hardware energies remain above the propagated energies; depth 2 converges faster classically but gives slightly worse deployed energy. For binarized MNIST generation, a 256-parameter circuit is trained using an observable-based MMD loss over 100,000 Adam steps. Truncated-propagator evaluation gives test $\mathrm{MMD}^2=(2.05\pm0.18)\times10^{-3}$, versus $(24.11\pm1.16)\times10^{-3}$ and $(2.51\pm0.22)\times10^{-3}$ for IQP baselines with 256 and 2,560 parameters. These scores are not hardware-sample evaluations. Hardware-generated images include recognizable digits and indistinct patterns. Stated limitations include expensive propagation of many generative-loss observables, approximate training, limited shallow-circuit expressivity, and hardware noise. Single-site statistics are reproduced more closely than pairwise correlations. The implemented ansatz is not strictly locally scrambling, so cited average-case truncation guarantees do not directly apply; general-purpose scalability remains open.

Little sign of AI-generated textIncrementalOverclaimsStrong assumptionsMissing comparison

Central claim may not hold: The abstract's claim of retaining classically intractable states or samples is unsupported by demonstrating 64-qubit circuits beyond convenient full-statevector simulation. The weight-truncation justification neither directly applies to the demonstrated non-locally-scrambling ansatz nor guarantees accuracy at parameters chosen by surrogate optimization. The truncated depth-2 VQE and generative objectives lack a reported independent accuracy assessment at the trained parameters, despite useful exact depth-1 and hardware demonstrations. The generative comparison uses different expectation-evaluation methods for the proposed model and IQP baselines without quantifying propagation bias, so its uncertainties do not establish that the reported advantage reflects modelling error alone.

Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.

Related papers

Symbolic Pauli Propagation for Gradient-Enabled Pre-Training of Quantum Circuits

Saverio Monaco, Jamal Slim, Florian Rehm +2 more

Quantum Machine Learning models typically require expensive on-chip training procedures and often lack efficient gradient estimation methods. By employing Pauli propagation, it is possible to derive a symbolic representation of observables as analytic functions of a circuit's parameters. Although the number of terms in such functional representations grows rapidly with circuit depth, suitable choices of ansatz and controlled truncations on Pauli weights and trigonometric degree components yield accurate yet tractable estimators of the target observables. With the right ansatz design, this approach can be extended to system sizes beyond the reach of classical statevector simulation, enabling scalable training for larger quantum systems. This also enables a form of classical pre-training through gradient-based optimization prior to deployment on quantum hardware. The proposed approach is demonstrated on the Variational Quantum Eigensolver for obtaining the ground state of the ANNNI spin model on 32 qubits, showing that accurate results can be achieved with a scalable and computationally efficient procedure.

quant-pharXiv:2512.16674•Dec 19, 2025•5 citations

Sample-based training of quantum generative models

Maria Demidik, Cenk Tüysüz, Michele Grossi +1 more

Quantum computers can efficiently sample from probability distributions that are believed to be classically intractable, providing a foundation for quantum generative modeling. However, practical training of such models remains challenging, as gradient evaluation via the parameter-shift rule scales linearly with the number of parameters and requires repeated expectation-value estimation under finite-shot noise. We introduce a training framework that extends the principle of contrastive divergence to quantum models. By deriving the circuit structure and providing a general recipe for constructing it, we obtain quantum circuits that generate the samples required for parameter updates, yielding constant scaling with respect to the cost of a forward pass, analogous to backpropagation in classical neural networks. Numerical results demonstrate that it attains comparable accuracy to likelihood-based optimization while requiring substantially fewer samples. The framework thereby establishes a scalable route to training expressive quantum generative models directly on quantum hardware.

quant-pharXiv:2511.11802•Nov 18, 2025•8 citations

Train on classical, deploy on quantum: scaling generative quantum machine learning to a thousand qubits

Erik Recio-Armengol, Shahnawaz Ahmed, Bowles, Joseph

We propose an approach to generative quantum machine learning that overcomes the fundamental scaling issues of variational quantum circuits. The core idea is to use a class of generative models based on instantaneous quantum polynomial circuits, which we show can be trained efficiently on classical hardware. Although training is classically efficient, sampling from these circuits is widely believed to be classically hard, and so computational advantages are possible when sampling from the trained model on quantum hardware. By combining our approach with a data-dependent parameter initialisation strategy, we do not encounter issues of barren plateaus and successfully circumvent the poor scaling of gradient estimation that plagues traditional approaches to quantum circuit optimisation. We investigate and evaluate our approach on a number of real and synthetic datasets, training models with up to one thousand qubits and hundreds of thousands of parameters. We find that the quantum models can successfully learn from high dimensional data, and perform surprisingly well compared to simple energy-based classical generative models trained with a similar amount of hyperparameter optimisation. Overall, our work demonstrates that a path to scalable quantum generative machine learning exists and can be investigated today at large scales.

arXiv:2503.02934•Mar 4, 2025•61 citations

Shallow instantaneous quantum polynomial-time circuits for generative modeling on noisy intermediate-scale quantum hardware

Oriol Balló-Gimbernat, Marcos Arroyo-Sánchez, Paula García-Molina +2 more

Generative modeling is one of the most promising applications of quantum machine learning, yet training and deploying Quantum Generative Models (QGMs) on near-term hardware remains effectively intractable due to prohibitive gradient estimation and implementation costs. We propose a resource-efficient approach based on shallow Instantaneous Quantum Polynomial-time (IQP) circuits that circumvents these bottlenecks by leveraging efficient classical training while retaining the guarantee of sampling hardness. To validate this approach, we formalize graph generation as a hierarchy of physical correlations, allowing us to map abstract data features, such as edge density and bipartiteness, directly to the quantum observables required to learn them. We validate our protocol through demonstrations both on real hardware (from $28$ to $153$ qubits) and simulations ($28$ qubits). Results show that while global structural features exhibit significant degradation beyond $91$ qubits, our models achieve high-precision reproduction of local correlations, even up to $153$ qubits. These findings establish shallow IQP circuits as a robust, scalable candidate for generative tasks on current Noisy Intermediate-Scale Quantum (NISQ) devices.

quant-pharXiv:2511.05267•Physical Review A•Nov 10, 2025•4 citations

Backpropagation scaling in parameterised quantum circuits

Joseph Bowles, David Wierichs, Chae-Yeun Park

The discovery of the backpropagation algorithm ranks among one of the most important moments in the history of machine learning, and has made possible the training of large-scale neural networks through its ability to compute gradients at roughly the same computational cost as model evaluation. Despite its importance, a similar backpropagation-like scaling for gradient evaluation of parameterised quantum circuits has remained elusive. Currently, the most popular method requires sampling from a number of circuits that scales with the number of circuit parameters, making training of large-scale quantum circuits prohibitively expensive in practice. Here we address this problem by introducing a class of structured circuits that are not known to be classically simulable and admit gradient estimation with significantly fewer circuits. In the simplest case – for which the parameters feed into commuting quantum gates – these circuits allow for fast estimation of the gradient, higher order partial derivatives and the Fisher information matrix. Moreover, specific families of parameterised circuits exist for which the scaling of gradient estimation is in line with classical backpropagation, and can thus be trained at scale. In a toy classification problem on 16 qubits, such circuits show competitive performance with other methods, while reducing the training cost by about two orders of magnitude.

arXiv:2306.14962•Quantum•Oct 2, 2025•43 citations

Optimization Methods for Quantum Machine Learning

Hamid D. Ismail

This chapter examines optimization as the central practical bottleneck in quantum machine learning (QML), emphasizing that the usefulness of parameterized quantum models depends not only on expressivity but also on whether they can be trained reliably under realistic conditions. It formulates QML training as a constrained, stochastic optimization problem on curved state manifolds, where parameters define unitary evolutions rather than unconstrained Euclidean weights. This geometric perspective is used to analyze the roles of parameterized quantum circuits, generator structure, circuit depth, and quantum Fisher information in shaping trainability and convergence behavior. This chapter develops the main classes of quantum loss functions and gradient-evaluation methods, including the parameter-shift rule, Hadamard-test-based approaches, and finite-shot estimation, and then surveys both gradient-based and gradient-free optimization methods relevant to hybrid quantum-classical training. Particular attention is given to major optimization pathologies, including barren plateaus, noise-induced flat landscapes, hyperparameter sensitivity, and parameter redundancy. The discussion also emphasizes hardware-aware optimization, accounting for finite shot budgets, noise, measurement overhead, and simulation strategies. Taken together, this chapter provides a systematic framework for understanding how geometry, stochastic measurement, ansatz design, optimizer choice, and hardware constraints jointly determine the practical limits of trainability in QML.

CRC Press eBooks•Nov 4, 2026•0 citations
Scaling parametrized quantum circuits to a larger number of qubits is a central challenge in Quantum Machine Learning: gradient estimation via the parameter-shift rule quickly becomes prohibitively expensive, while gradient-free optimizers offer limited convergence guarantees. We apply the Pauli propagation framework to a broad class of Quantum Machine Learning problems in which the loss is expressed as a function of expectation values of the circuit's observables, while keeping the relevant quantity of interest, such as the quantum state or its samples, classically intractable to compute. By propagating the observables backward through the circuit, we obtain an explicit parametric expression for the loss that can be minimized entirely on classical hardware. We also present complementary strategies to make propagation tractable at large scale, including weight-based and degree-based truncations and pruning heuristics. We demonstrate the approach on two benchmarks run on 64-qubit circuits: Variational Quantum Eigensolver on a two-dimensional Ising model, and training of a discrete-type generative model for binary images. In both cases, the classically pre-trained parameters are deployed on the IBM quantum processor ibm_berlin. For generative learning, we additionally show that the circuits considered here require substantially fewer parametrized gates than IQP-style ansätze, reducing transpilation overhead and making deployment on current quantum hardware feasible.
The paper asks whether large, shallow parametrized quantum circuits can be trained classically using observable-based losses, then deployed on quantum hardware. It backward-propagates observables through the circuit to construct explicit trigonometric functions of the parameters, optimized with classical gradients. Pauli-weight and trigonometric-degree cutoffs control expression growth; two exact pruning rules remove branches that cannot contribute to the initial-state expectation value.
Demonstrations use 64-qubit circuits on an 8×88\times88×8 grid, with deployment on IBM’s ibm_berlin. For VQE on the two-dimensional transverse-field Ising model at (J,h)=(1,2)(J,h)=(1,2)(J,h)=(1,2), depth-1 and depth-2 ansätze have 256 and 384 parameters. Training uses 1,000 Adam steps, with exact propagation at depth 1 and weight cutoff wcut=12w_{\mathrm{cut}}=12wcut​=12 at depth 2. Hardware energies remain above the propagated energies; depth 2 converges faster classically but gives slightly worse deployed energy.
For binarized MNIST generation, a 256-parameter circuit is trained using an observable-based MMD loss over 100,000 Adam steps. Truncated-propagator evaluation gives test MMD2=(2.05±0.18)×10−3\mathrm{MMD}^2=(2.05\pm0.18)\times10^{-3}MMD2=(2.05±0.18)×10−3, versus (24.11±1.16)×10−3(24.11\pm1.16)\times10^{-3}(24.11±1.16)×10−3 and (2.51±0.22)×10−3(2.51\pm0.22)\times10^{-3}(2.51±0.22)×10−3 for IQP baselines with 256 and 2,560 parameters. These scores are not hardware-sample evaluations. Hardware-generated images include recognizable digits and indistinct patterns.
Stated limitations include expensive propagation of many generative-loss observables, approximate training, limited shallow-circuit expressivity, and hardware noise. Single-site statistics are reproduced more closely than pairwise correlations. The implemented ansatz is not strictly locally scrambling, so cited average-case truncation guarantees do not directly apply; general-purpose scalability remains open.
Quantum Machine Learning models typically require expensive on-chip training procedures and often lack efficient gradient estimation methods. By employing Pauli propagation, it is possible to derive a symbolic representation of observables as analytic functions of a circuit's parameters. Although the number of terms in such functional representations grows rapidly with circuit depth, suitable choices of ansatz and controlled truncations on Pauli weights and trigonometric degree components yield accurate yet tractable estimators of the target observables. With the right ansatz design, this approach can be extended to system sizes beyond the reach of classical statevector simulation, enabling scalable training for larger quantum systems. This also enables a form of classical pre-training through gradient-based optimization prior to deployment on quantum hardware. The proposed approach is demonstrated on the Variational Quantum Eigensolver for obtaining the ground state of the ANNNI spin model on 32 qubits, showing that accurate results can be achieved with a scalable and computationally efficient procedure.
Quantum computers can efficiently sample from probability distributions that are believed to be classically intractable, providing a foundation for quantum generative modeling. However, practical training of such models remains challenging, as gradient evaluation via the parameter-shift rule scales linearly with the number of parameters and requires repeated expectation-value estimation under finite-shot noise. We introduce a training framework that extends the principle of contrastive divergence to quantum models. By deriving the circuit structure and providing a general recipe for constructing it, we obtain quantum circuits that generate the samples required for parameter updates, yielding constant scaling with respect to the cost of a forward pass, analogous to backpropagation in classical neural networks. Numerical results demonstrate that it attains comparable accuracy to likelihood-based optimization while requiring substantially fewer samples. The framework thereby establishes a scalable route to training expressive quantum generative models directly on quantum hardware.
We propose an approach to generative quantum machine learning that overcomes the fundamental scaling issues of variational quantum circuits. The core idea is to use a class of generative models based on instantaneous quantum polynomial circuits, which we show can be trained efficiently on classical hardware. Although training is classically efficient, sampling from these circuits is widely believed to be classically hard, and so computational advantages are possible when sampling from the trained model on quantum hardware. By combining our approach with a data-dependent parameter initialisation strategy, we do not encounter issues of barren plateaus and successfully circumvent the poor scaling of gradient estimation that plagues traditional approaches to quantum circuit optimisation. We investigate and evaluate our approach on a number of real and synthetic datasets, training models with up to one thousand qubits and hundreds of thousands of parameters. We find that the quantum models can successfully learn from high dimensional data, and perform surprisingly well compared to simple energy-based classical generative models trained with a similar amount of hyperparameter optimisation. Overall, our work demonstrates that a path to scalable quantum generative machine learning exists and can be investigated today at large scales.
Generative modeling is one of the most promising applications of quantum machine learning, yet training and deploying Quantum Generative Models (QGMs) on near-term hardware remains effectively intractable due to prohibitive gradient estimation and implementation costs. We propose a resource-efficient approach based on shallow Instantaneous Quantum Polynomial-time (IQP) circuits that circumvents these bottlenecks by leveraging efficient classical training while retaining the guarantee of sampling hardness. To validate this approach, we formalize graph generation as a hierarchy of physical correlations, allowing us to map abstract data features, such as edge density and bipartiteness, directly to the quantum observables required to learn them. We validate our protocol through demonstrations both on real hardware (from 282828 to 153153153 qubits) and simulations (282828 qubits). Results show that while global structural features exhibit significant degradation beyond 919191 qubits, our models achieve high-precision reproduction of local correlations, even up to 153153153 qubits. These findings establish shallow IQP circuits as a robust, scalable candidate for generative tasks on current Noisy Intermediate-Scale Quantum (NISQ) devices.
The discovery of the backpropagation algorithm ranks among one of the most important moments in the history of machine learning, and has made possible the training of large-scale neural networks through its ability to compute gradients at roughly the same computational cost as model evaluation. Despite its importance, a similar backpropagation-like scaling for gradient evaluation of parameterised quantum circuits has remained elusive. Currently, the most popular method requires sampling from a number of circuits that scales with the number of circuit parameters, making training of large-scale quantum circuits prohibitively expensive in practice. Here we address this problem by introducing a class of structured circuits that are not known to be classically simulable and admit gradient estimation with significantly fewer circuits. In the simplest case – for which the parameters feed into commuting quantum gates – these circuits allow for fast estimation of the gradient, higher order partial derivatives and the Fisher information matrix. Moreover, specific families of parameterised circuits exist for which the scaling of gradient estimation is in line with classical backpropagation, and can thus be trained at scale. In a toy classification problem on 16 qubits, such circuits show competitive performance with other methods, while reducing the training cost by about two orders of magnitude.
This chapter examines optimization as the central practical bottleneck in quantum machine learning (QML), emphasizing that the usefulness of parameterized quantum models depends not only on expressivity but also on whether they can be trained reliably under realistic conditions. It formulates QML training as a constrained, stochastic optimization problem on curved state manifolds, where parameters define unitary evolutions rather than unconstrained Euclidean weights. This geometric perspective is used to analyze the roles of parameterized quantum circuits, generator structure, circuit depth, and quantum Fisher information in shaping trainability and convergence behavior. This chapter develops the main classes of quantum loss functions and gradient-evaluation methods, including the parameter-shift rule, Hadamard-test-based approaches, and finite-shot estimation, and then surveys both gradient-based and gradient-free optimization methods relevant to hybrid quantum-classical training. Particular attention is given to major optimization pathologies, including barren plateaus, noise-induced flat landscapes, hyperparameter sensitivity, and parameter redundancy. The discussion also emphasizes hardware-aware optimization, accounting for finite shot budgets, noise, measurement overhead, and simulation strategies. Taken together, this chapter provides a systematic framework for understanding how geometry, stochastic measurement, ansatz design, optimizer choice, and hardware constraints jointly determine the practical limits of trainability in QML.