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:2503.02934

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

Erik Recio-Armengol, Shahnawaz Ahmed, Bowles, Joseph

Published Mar 4, 2025• 61 citations
arXivHTMLPDF

Abstract

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.

DOI
10.48550/arxiv.2503.02934

Related papers

Protocols for classically training quantum generative models on probability distributions

Sachin Kasture, Oleksandr Kyriienko, Vincent E. Elfving

Quantum generative modeling (QGM) relies on preparing quantum states and generating samples from these states as hidden---or known---probability distributions. As distributions from some classes of quantum states (circuits) are inherently hard to sample classically, QGM represents an excellent test bed for quantum supremacy experiments. Furthermore, generative tasks are increasingly relevant for industrial machine learning applications, and thus QGM is a strong candidate for demonstrating a practical quantum advantage. However, this requires that quantum circuits are trained to represent industrially relevant distributions, and the corresponding training stage has an extensive training cost for current quantum hardware in practice. In this work, we propose protocols for classical training of QGMs based on circuits of the specific type that admit an efficient gradient computation, while remaining hard to sample. In particular, we consider instantaneous quantum polynomial (IQP) circuits and their extensions. Showing their classical simulability in terms of the time complexity, sparsity, and anticoncentration properties, we develop a classically tractable way of simulating their output probability distributions, allowing classical training to a target probability distribution. The corresponding quantum sampling from IQPs can be performed efficiently, unlike when using classical sampling. We numerically demonstrate the end-to-end training of IQP circuits using probability distributions for up to 30 qubits on a regular desktop computer. When applied to industrially relevant distributions this combination of classical training with quantum sampling represents an avenue for reaching advantage in the noisy intermediate-scale quantum (NISQ) era.

arXiv:2210.13442•Physical Review A•Oct 5, 2023•19 citations

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

Saverio Monaco, Jamal Slim, Ran Xue +2 more

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.

quant-pharXiv:2610.32571•Oct 26, 2026•0 citations

"Train classical, deploy quantum" requires rethinking generalization

Snehal Raj, Natansh Mathur, Alejandro Perdomo-Ortiz

Generative models have become central across science and industry, from image and text synthesis to the design of molecules and materials. Quantum generative models are considered one of the most promising applications for quantum computers, since a quantum circuit naturally produces samples from the distribution it encodes, and for suitable circuits that distribution is believed to be hard for any classical computer to reproduce. A leading strategy trains these models on a classical computer and reserves the quantum device for generating samples at deployment. This is possible when the training loss can be evaluated on a classical computer. A prime example is the maximum mean discrepancy (MMD$^2$), a moment-matching loss that compares the model and the data through their Pauli-$Z$ correlations. Research so far has asked whether such models can be trained and whether their sampling is hard; whether minimizing such an objective yields a model that \emph{generalizes}, rather than one that merely reproduces the training statistics, remains poorly understood. We benchmark thirteen quantum and classical generative models by direct sampling on two application-inspired datasets: first a cardinality-constrained dataset at up to $30$ qubits and second a dataset of genomic single-nucleotide variants, whose valid set is the observed data. Models that converge the loss to the same value differ widely in how much of the unseen valid set they cover. These results indicate that a converged moment-matching loss is not a reliable measure of generalization, and that a train-classical, deploy-quantum workflow has to measure generalization by sampling the trained model, a step that at the sizes of interest is believed to require the quantum device.

quant-pharXiv:2608.31117•Sep 1, 2026•3 citations

Hierarchical Learning for Quantum ML: Novel Training Technique for Large-Scale Variational Quantum Circuits

Hrant Gharibyan, Su, Vincent, Hayk Tepanyan

We present hierarchical learning, a novel variational architecture for efficient training of large-scale variational quantum circuits. We test and benchmark our technique for distribution loading with quantum circuit born machines (QCBMs). With QCBMs, probability distributions are loaded into the squared amplitudes of computational basis vectors represented by bitstrings. Our key insight is to take advantage of the fact that the most significant (qu)bits have a greater effect on the final distribution and can be learned first. One can think of it as a generalization of layerwise learning, where some parameters of the variational circuit are learned first to prevent the phenomena of barren plateaus. We briefly review adjoint methods for computing the gradient, in particular for loss functions that are not expectation values of observables. We first compare the role of connectivity in the variational ansatz for the task of loading a Gaussian distribution on nine qubits, finding that 2D connectivity greatly outperforms qubits arranged on a line. Based on our observations, we then implement this strategy on large-scale numerical experiments with GPUs, training a QCBM to reproduce a 3-dimensional multivariate Gaussian distribution on 27 qubits up to $\sim4\%$ total variation distance. Though barren plateau arguments do not strictly apply here due to the objective function not being tied to an observable, this is to our knowledge the first practical demonstration of variational learning on large numbers of qubits. We also demonstrate hierarchical learning as a resource-efficient way to load distributions for existing quantum hardware (IBM's 7 and 27 qubit devices) in tandem with Fire Opal optimizations.

arXiv:2311.12929•Nov 21, 2023•8 citations

Synthetic data generation on noisy quantum computers

Michael Krebsbach

Generative machine learning aims to model probability distributions from sample data. Quantum computers are widely believed to efficiently sample from certain probability distributions that are intractable for classical computers, making quantum circuits promising candidates for generative machine learning. This dissertation investigates quantum generative machine learning models, with a particular focus on Synthetic Data Generation (SDG) and on the Quantum Circuit Born Machine, an algorithm that can be implemented on current quantum hardware. The first part of this dissertation studies SDG algorithms in noiseless simulations. Most of these algorithms rely on an implicit design choice that has not previously been examined: the binary encoding that maps bitstring samples generated by a quantum circuit to real-valued data points. We show that the training performance depends on this choice and analyze the interplay between data structure, circuit ansatz, and encoding. For continuous target distributions, the reflected binary Gray code is particularly effective because it maps similar data points to similar bitstrings and thus preserves the structure of the data. This provides a useful inductive bias that leads, on average, to faster convergence and better synthetic data. We further investigate extensions of SDG algorithms. Contrary to previous findings, we show that quantum models can also learn conditional probability distributions with quality comparable to that of independently trained models. We develop two sample-efficient strategies that enable conditional models to perform classification without additional training. This yields a unified model with both generative and discriminative capabilities. The second part of this dissertation addresses the implementation of such models on real quantum hardware. While current devices can train small-scale models with reasonable resource, larger-scale implementations remain limited by noise. Understanding and modeling this noise is therefore essential for progress. First, we develop a physically motivated noise model for superconducting transmon systems based on Hamiltonian simulation. We show that frequency collisions between energy levels of neighboring transmons can induce crosstalk and may drive the system out of the computational subspace. This model provides a basis for understanding hardware-level error mechanisms and for developing taylored mitigation or correction strategies. Second, we investigate two-qubit error using Long-Sequence Quantum Process Tomography (LS-QPT), assuming well-characterized single-qubit gates. Compared with standard Quantum Process Tomography, LS-QPT reduces the measurement cost by amplifying noise through germ circuits. We show that germ circuits that ideally implement the identity operation amplify a larger subspace of noise processes than other types of germ circuits. We also show that the precision of LS-QPT is fundamentally limited by unlearnable degrees of freedom in the noise model that cannot be amplified. Despite this limitation, the LS-QPT method reduces the total circuit depth required for process tomography by an order of magnitude.

•Jan 1, 2026•0 citations

Training Theory of Variational Quantum Machine Learning

Kaining Zhang

Recent advancements in machine learning have revolutionised research across various fields. Despite their success, conventional learning techniques are hindered by their significant computational resources and energy requirements. Prompted by recent experimental breakthroughs in quantum computing, variational quantum machine learning (QML) – machine learning integrated with variational quantum circuits (VQCs) – has emerged as a promising alternative. Nonetheless, the theoretical framework underpinning the advantages of variational QML is still rudimentary. Specifically, the training of VQCs faces several challenges, such as the barren plateau problem, where the gradient diminishes exponentially with an increasing qubits. A related issue arises in variational QML training, where the convergence rate is exponentially small. In this thesis, we present theoretically guaranteed solutions to these challenges. First, we construct innovative circuit architectures to address the vanishing gradient problem in deep VQCs. We propose quantum controlled-layer and quantum ResNet structures, demonstrating that the expected gradient norm's lower bound is unaffected by the increase in qubits and circuit depth. Next, we introduce an initialization strategy to mitigate the vanishing gradient issue in general deep quantum circuits. We prove that Gaussian-initialized parameters ensure the gradient norm's decay rate remains inversely polynomial despite the increase in qubit numbers and circuit depth. Finally, we propose a novel and effective theory for analysing the training of quantum neural networks with moderate depths. We prove that, under certain randomness conditions in the circuits and datasets, training converges linearly with a rate inversely proportional to the dataset size. Our approach surpasses previous results, achieving exponentially larger convergence rates with modest depth, or conversely, requiring exponentially less depth for equivalent rates.

•Jan 1, 2023•0 citations
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.
Quantum generative modeling (QGM) relies on preparing quantum states and generating samples from these states as hidden---or known---probability distributions. As distributions from some classes of quantum states (circuits) are inherently hard to sample classically, QGM represents an excellent test bed for quantum supremacy experiments. Furthermore, generative tasks are increasingly relevant for industrial machine learning applications, and thus QGM is a strong candidate for demonstrating a practical quantum advantage. However, this requires that quantum circuits are trained to represent industrially relevant distributions, and the corresponding training stage has an extensive training cost for current quantum hardware in practice. In this work, we propose protocols for classical training of QGMs based on circuits of the specific type that admit an efficient gradient computation, while remaining hard to sample. In particular, we consider instantaneous quantum polynomial (IQP) circuits and their extensions. Showing their classical simulability in terms of the time complexity, sparsity, and anticoncentration properties, we develop a classically tractable way of simulating their output probability distributions, allowing classical training to a target probability distribution. The corresponding quantum sampling from IQPs can be performed efficiently, unlike when using classical sampling. We numerically demonstrate the end-to-end training of IQP circuits using probability distributions for up to 30 qubits on a regular desktop computer. When applied to industrially relevant distributions this combination of classical training with quantum sampling represents an avenue for reaching advantage in the noisy intermediate-scale quantum (NISQ) era.
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.
Generative models have become central across science and industry, from image and text synthesis to the design of molecules and materials. Quantum generative models are considered one of the most promising applications for quantum computers, since a quantum circuit naturally produces samples from the distribution it encodes, and for suitable circuits that distribution is believed to be hard for any classical computer to reproduce. A leading strategy trains these models on a classical computer and reserves the quantum device for generating samples at deployment. This is possible when the training loss can be evaluated on a classical computer. A prime example is the maximum mean discrepancy (MMD2^22), a moment-matching loss that compares the model and the data through their Pauli-ZZZ correlations. Research so far has asked whether such models can be trained and whether their sampling is hard; whether minimizing such an objective yields a model that \emph{generalizes}, rather than one that merely reproduces the training statistics, remains poorly understood. We benchmark thirteen quantum and classical generative models by direct sampling on two application-inspired datasets: first a cardinality-constrained dataset at up to 303030 qubits and second a dataset of genomic single-nucleotide variants, whose valid set is the observed data. Models that converge the loss to the same value differ widely in how much of the unseen valid set they cover. These results indicate that a converged moment-matching loss is not a reliable measure of generalization, and that a train-classical, deploy-quantum workflow has to measure generalization by sampling the trained model, a step that at the sizes of interest is believed to require the quantum device.
We present hierarchical learning, a novel variational architecture for efficient training of large-scale variational quantum circuits. We test and benchmark our technique for distribution loading with quantum circuit born machines (QCBMs). With QCBMs, probability distributions are loaded into the squared amplitudes of computational basis vectors represented by bitstrings. Our key insight is to take advantage of the fact that the most significant (qu)bits have a greater effect on the final distribution and can be learned first. One can think of it as a generalization of layerwise learning, where some parameters of the variational circuit are learned first to prevent the phenomena of barren plateaus. We briefly review adjoint methods for computing the gradient, in particular for loss functions that are not expectation values of observables. We first compare the role of connectivity in the variational ansatz for the task of loading a Gaussian distribution on nine qubits, finding that 2D connectivity greatly outperforms qubits arranged on a line. Based on our observations, we then implement this strategy on large-scale numerical experiments with GPUs, training a QCBM to reproduce a 3-dimensional multivariate Gaussian distribution on 27 qubits up to ∼4%\sim4\%∼4% total variation distance. Though barren plateau arguments do not strictly apply here due to the objective function not being tied to an observable, this is to our knowledge the first practical demonstration of variational learning on large numbers of qubits. We also demonstrate hierarchical learning as a resource-efficient way to load distributions for existing quantum hardware (IBM's 7 and 27 qubit devices) in tandem with Fire Opal optimizations.
Generative machine learning aims to model probability distributions from sample data. Quantum computers are widely believed to efficiently sample from certain probability distributions that are intractable for classical computers, making quantum circuits promising candidates for generative machine learning. This dissertation investigates quantum generative machine learning models, with a particular focus on Synthetic Data Generation (SDG) and on the Quantum Circuit Born Machine, an algorithm that can be implemented on current quantum hardware. The first part of this dissertation studies SDG algorithms in noiseless simulations. Most of these algorithms rely on an implicit design choice that has not previously been examined: the binary encoding that maps bitstring samples generated by a quantum circuit to real-valued data points. We show that the training performance depends on this choice and analyze the interplay between data structure, circuit ansatz, and encoding. For continuous target distributions, the reflected binary Gray code is particularly effective because it maps similar data points to similar bitstrings and thus preserves the structure of the data. This provides a useful inductive bias that leads, on average, to faster convergence and better synthetic data. We further investigate extensions of SDG algorithms. Contrary to previous findings, we show that quantum models can also learn conditional probability distributions with quality comparable to that of independently trained models. We develop two sample-efficient strategies that enable conditional models to perform classification without additional training. This yields a unified model with both generative and discriminative capabilities. The second part of this dissertation addresses the implementation of such models on real quantum hardware. While current devices can train small-scale models with reasonable resource, larger-scale implementations remain limited by noise. Understanding and modeling this noise is therefore essential for progress. First, we develop a physically motivated noise model for superconducting transmon systems based on Hamiltonian simulation. We show that frequency collisions between energy levels of neighboring transmons can induce crosstalk and may drive the system out of the computational subspace. This model provides a basis for understanding hardware-level error mechanisms and for developing taylored mitigation or correction strategies. Second, we investigate two-qubit error using Long-Sequence Quantum Process Tomography (LS-QPT), assuming well-characterized single-qubit gates. Compared with standard Quantum Process Tomography, LS-QPT reduces the measurement cost by amplifying noise through germ circuits. We show that germ circuits that ideally implement the identity operation amplify a larger subspace of noise processes than other types of germ circuits. We also show that the precision of LS-QPT is fundamentally limited by unlearnable degrees of freedom in the noise model that cannot be amplified. Despite this limitation, the LS-QPT method reduces the total circuit depth required for process tomography by an order of magnitude.
Recent advancements in machine learning have revolutionised research across various fields. Despite their success, conventional learning techniques are hindered by their significant computational resources and energy requirements. Prompted by recent experimental breakthroughs in quantum computing, variational quantum machine learning (QML) – machine learning integrated with variational quantum circuits (VQCs) – has emerged as a promising alternative. Nonetheless, the theoretical framework underpinning the advantages of variational QML is still rudimentary. Specifically, the training of VQCs faces several challenges, such as the barren plateau problem, where the gradient diminishes exponentially with an increasing qubits. A related issue arises in variational QML training, where the convergence rate is exponentially small. In this thesis, we present theoretically guaranteed solutions to these challenges. First, we construct innovative circuit architectures to address the vanishing gradient problem in deep VQCs. We propose quantum controlled-layer and quantum ResNet structures, demonstrating that the expected gradient norm's lower bound is unaffected by the increase in qubits and circuit depth. Next, we introduce an initialization strategy to mitigate the vanishing gradient issue in general deep quantum circuits. We prove that Gaussian-initialized parameters ensure the gradient norm's decay rate remains inversely polynomial despite the increase in qubit numbers and circuit depth. Finally, we propose a novel and effective theory for analysing the training of quantum neural networks with moderate depths. We prove that, under certain randomness conditions in the circuits and datasets, training converges linearly with a rate inversely proportional to the dataset size. Our approach surpasses previous results, achieving exponentially larger convergence rates with modest depth, or conversely, requiring exponentially less depth for equivalent rates.