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:2510.13210 · Journal of the Physical Society of Japan

Performance Evaluation of Ising and QUBO Variable Encodings in Boltzmann Machine Learning

Yasushi Hasegawa, Masayuki Ohzeki

Published Sep 29, 2026• 0 citations
arXivHTMLPDF

Abstract

We compare Ising ([Formula: see text]) and QUBO ([Formula: see text]) encodings for Boltzmann machine learning under controlled protocols that fix the sampler, optimizer, and learning-rate design within each comparison. Exploiting the identity that the Fisher information matrix (FIM) equals the covariance of sufficient statistics, we visualize empirical moments from model samples and reveal systematic, representation-dependent differences. QUBO induces larger cross terms between first- and second-order statistics, creating more small-eigenvalue directions in the FIM and lowering spectral entropy. This ill-conditioning explains slower convergence under stochastic gradient descent (SGD). In contrast, full-FIM natural gradient descent (NGD), which rescales updates by the FIM metric, achieves similar convergence across encodings, whereas diagonal-FIM approximation can reintroduce representation-dependent differences. Practically, for SGD-based training, the Ising encoding provides more isotropic curvature and faster convergence; for QUBO, centering/scaling or NGD-style preconditioning mitigates curvature pathologies. These results clarify how representation shapes information geometry and finite-time learning dynamics in Boltzmann machines and yield actionable guidelines for variable encoding and preprocessing.

DOI
10.7566/jpsj.95.104006

Related papers

Evolved quantum Boltzmann machines

Michele Minervini, Dhrumil Patel, Mark M. Wilde

We introduce evolved quantum Boltzmann machines as a variational ansatz for quantum optimization and learning tasks. Given two parameterized Hamiltonians $G(θ)$ and $H(ϕ)$, an evolved quantum Boltzmann machine consists of preparing a thermal state of the first Hamiltonian $G(θ)$ followed by unitary evolution according to the second Hamiltonian $H(ϕ)$. Alternatively, one can think of it as first realizing imaginary time evolution according to $G(θ)$ followed by real time evolution according to $H(ϕ)$. After defining this ansatz, we provide analytical expressions for the gradient vector and illustrate their application in ground-state energy estimation and generative modeling, showing how the gradient for these tasks can be estimated by means of quantum algorithms that involve classical sampling, Hamiltonian simulation, and the Hadamard test. We also establish analytical expressions for the Fisher-Bures, Wigner-Yanase, and Kubo-Mori information matrix elements of evolved quantum Boltzmann machines, as well as quantum algorithms for estimating each of them, which leads to at least three different general natural gradient descent algorithms based on this ansatz. Along the way, we establish a broad generalization of the main result of [Luo, Proc. Am. Math. Soc. 132, 885 (2004)], proving that the Fisher-Bures and Wigner-Yanase information matrices of general parameterized families of states differ by no more than a factor of two in the matrix (Loewner) order, making them essentially interchangeable for training when using natural gradient descent.

arXiv:2501.03367•Physical Review A•Jan 21, 2026•15 citations

Natural gradient and parameter estimation for quantum Boltzmann machines

Dhrumil Patel, Mark M. Wilde

Thermal states play a fundamental role in various areas of physics, and they are becoming increasingly important in quantum information science, with applications related to semidefinite programming, quantum Boltzmann machine learning, Hamiltonian learning, and the related task of estimating the parameters of a Hamiltonian. Here we establish formulas underlying the basic geometry of parameterized thermal states, and we delineate quantum algorithms for estimating the values of these formulas. More specifically, we establish formulas for the Fisher--Bures and Kubo--Mori information matrices of parameterized thermal states, and our quantum algorithms for estimating their matrix elements involve a combination of classical sampling, Hamiltonian simulation, and the Hadamard test. These results have applications in developing a natural gradient descent algorithm for quantum Boltzmann machine learning, which takes into account the geometry of thermal states, and in establishing fundamental limitations on the ability to estimate the parameters of a Hamiltonian, when given access to thermal-state samples. For the latter task, and for the special case of estimating a single parameter, we sketch an algorithm that realizes a measurement that is asymptotically optimal for the estimation task. We finally stress that the natural gradient descent algorithm developed here can be used for any machine learning problem that employs the quantum Boltzmann machine ansatz.

arXiv:2410.24058•Physical Review A•Nov 12, 2025•18 citations

Fisher information, training and bias in Fourier regression models

Lorenzo Pastori, Veronika Eyring, Mierk Schwabe

Abstract Motivated by the growing interest in quantum machine learning, in particular quantum neural networks (QNNs), we study how recently introduced evaluation metrics based on the Fisher information matrix (FIM) are effective for predicting their training and prediction performance. We exploit the equivalence between a broad class of QNNs and Fourier models, and study the interplay between the effective dimension (ED) and the bias of a model towards a given task, investigating how these affect the model’s training and performance. We show that for a model that is completely agnostic, or unbiased, towards the function to be learned, a higher ED likely results in a better trainability and performance. On the other hand, for models that are biased towards the function to be learned a lower ED is likely beneficial during training. To obtain these results, we derive an analytical expression of the FIM for Fourier models and identify the features controlling a model’s ED. This allows us to construct models with tunable ED and bias, and to compare their training. We furthermore introduce a tensor network representation of the considered Fourier models, which could be a tool of independent interest for the analysis of QNN models. Overall, these findings provide an explicit example of the interplay between geometrical properties, model-task alignment and training, which are relevant for the broader machine learning community.

arXiv:2510.06945•Machine Learning Science and Technology•Jul 23, 2026•3 citations

Multi-Mode Quantum Annealing for Generative Representation Learning with Boltzmann Priors

Gilhan Kim, Daniel K. Park

Energy-based models provide a natural bridge between statistical physics and machine learning by representing data through structured energy landscapes. Boltzmann machines are a particularly compelling class of such models for capturing complex interactions among latent variables, but their use in modern generative learning has been limited by the classical intractability of sampling from general (non-restricted) Boltzmann distributions. Here we develop a quantum-annealing-based framework that enables variational autoencoders with general Boltzmann priors. The framework employs three complementary annealing modes tailored to different stages of learning and deployment: diabatic quantum annealing provides unbiased Boltzmann samples for efficient training, slower annealing concentrates samples near low-energy configurations of the learned prior for unconditional generation, and conditional annealing with external fields steers the learned energy landscape toward attribute-specific regions for conditional generation and semantic editing. Using up to 2000 qubits on a D-Wave Advantage2 processor, we demonstrate stable training and high-quality generation on MNIST, Fashion-MNIST, and CelebA, achieving faster convergence and lower reconstruction loss than a Gaussian-prior VAE with the same encoder-decoder architecture. Beyond generation, the learned energy function provides out-of-distribution detection signals that add discriminative power beyond reconstruction loss. We demonstrate that these scores separate in-distribution samples from held-out digit classes in one-class MNIST experiments and improve the detection of market regime shifts in financial data. These results establish quantum annealing as a practical and controllable physical mechanism for energy-based representation learning and generative modeling beyond the reach of tractable classical approaches.

quant-pharXiv:2604.00919•Apr 2, 2026•1 citation

A cross-domain empirical benchmark of quantum-inspired and classical optimization algorithms for machine learning

Görkem Yılmaz

Abstract Quantum-inspired (QI) optimizers are increasingly proposed as alternatives to classical gradient-based training, yet are frequently benchmarked through classical stand-ins rather than themselves. Across tabular, vision, and language tasks we evaluate three classes: direct implementations of published update rules (Spall’s SPSA; quantum-behaved particle swarm optimization, QPSO, adapted to stochastic minibatches), a documented classical analog of quantum natural gradient (an NGD-style diagonal second-moment preconditioner), and the hyperparameter-control configurations under which such methods are often reported. The gap between these classes is our central result: on the tabular benchmark the hyperparameter control trains with the classical baselines while the direct algorithms fall $$\approx $$ 10–24 points below it, and on CIFAR-10/ResNet18, where v1’s “SPSA” arm was a learning-rate-scaled Adam, Adam itself stands $$\approx $$ 59 points above genuine SPSA. The QPSO adaptation stays at chance on LSTM tasks a legacy “QPSO” proxy solves at 79–81%, at 18–38 $$\times $$ Adam’s cost there and up to 74 $$\times $$ on the small CV model. A seed-paired $$2\times 2$$ factorial on RoBERTa identifies the smaller learning rate as the dominant observed component of the earlier “QNG advantage” ( $$+0.50$$ points at the dataset level); with only three datasets, modified Hartung–Knapp uncertainty spans zero. The configured NGD arm shows no detectable advantage over a learning-rate-matched AdamW (descriptive pooled $$p{=}0.23$$ ) at 6– $$7\%$$ extra cost, and on ResNet18 its trained endpoints have larger dominant-curvature estimates than the Adam family’s. Headline comparisons use five seeds, paired tests, Holm correction, and stated detection limits. We conclude with an auditing taxonomy (direct algorithm / analog / hyperparameter control): within the arms, protocols, and scales evaluated, no advantage attributable to a QI mechanism is detected after these controls.

Quantum Machine Intelligence•Sep 11, 2026•1 citation

Simultaneous Perturbation Stochastic Approximation of the Quantum Fisher Information

Julien Gacon, Christa Zoufal, Giuseppe Carleo +1 more

The Quantum Fisher Information matrix (QFIM) is a central metric in promising algorithms, such as Quantum Natural Gradient Descent and Variational Quantum Imaginary Time Evolution. Computing the full QFIM for a model with d parameters, however, is computationally expensive and generally requires O ( d 2 ) function evaluations. To remedy these increasing costs in high-dimensional parameter spaces, we propose using simultaneous perturbation stochastic approximation techniques to approximate the QFIM at a constant cost. We present the resulting algorithm and successfully apply it to prepare Hamiltonian ground states and train Variational Quantum Boltzmann Machines.

arXiv:2103.09232•Quantum•Oct 20, 2021•115 citations
We compare Ising ([Formula: see text]) and QUBO ([Formula: see text]) encodings for Boltzmann machine learning under controlled protocols that fix the sampler, optimizer, and learning-rate design within each comparison. Exploiting the identity that the Fisher information matrix (FIM) equals the covariance of sufficient statistics, we visualize empirical moments from model samples and reveal systematic, representation-dependent differences. QUBO induces larger cross terms between first- and second-order statistics, creating more small-eigenvalue directions in the FIM and lowering spectral entropy. This ill-conditioning explains slower convergence under stochastic gradient descent (SGD). In contrast, full-FIM natural gradient descent (NGD), which rescales updates by the FIM metric, achieves similar convergence across encodings, whereas diagonal-FIM approximation can reintroduce representation-dependent differences. Practically, for SGD-based training, the Ising encoding provides more isotropic curvature and faster convergence; for QUBO, centering/scaling or NGD-style preconditioning mitigates curvature pathologies. These results clarify how representation shapes information geometry and finite-time learning dynamics in Boltzmann machines and yield actionable guidelines for variable encoding and preprocessing.
We introduce evolved quantum Boltzmann machines as a variational ansatz for quantum optimization and learning tasks. Given two parameterized Hamiltonians G(θ)G(θ)G(θ) and H(ϕ)H(ϕ)H(ϕ), an evolved quantum Boltzmann machine consists of preparing a thermal state of the first Hamiltonian G(θ)G(θ)G(θ) followed by unitary evolution according to the second Hamiltonian H(ϕ)H(ϕ)H(ϕ). Alternatively, one can think of it as first realizing imaginary time evolution according to G(θ)G(θ)G(θ) followed by real time evolution according to H(ϕ)H(ϕ)H(ϕ). After defining this ansatz, we provide analytical expressions for the gradient vector and illustrate their application in ground-state energy estimation and generative modeling, showing how the gradient for these tasks can be estimated by means of quantum algorithms that involve classical sampling, Hamiltonian simulation, and the Hadamard test. We also establish analytical expressions for the Fisher-Bures, Wigner-Yanase, and Kubo-Mori information matrix elements of evolved quantum Boltzmann machines, as well as quantum algorithms for estimating each of them, which leads to at least three different general natural gradient descent algorithms based on this ansatz. Along the way, we establish a broad generalization of the main result of [Luo, Proc. Am. Math. Soc. 132, 885 (2004)], proving that the Fisher-Bures and Wigner-Yanase information matrices of general parameterized families of states differ by no more than a factor of two in the matrix (Loewner) order, making them essentially interchangeable for training when using natural gradient descent.
Thermal states play a fundamental role in various areas of physics, and they are becoming increasingly important in quantum information science, with applications related to semidefinite programming, quantum Boltzmann machine learning, Hamiltonian learning, and the related task of estimating the parameters of a Hamiltonian. Here we establish formulas underlying the basic geometry of parameterized thermal states, and we delineate quantum algorithms for estimating the values of these formulas. More specifically, we establish formulas for the Fisher--Bures and Kubo--Mori information matrices of parameterized thermal states, and our quantum algorithms for estimating their matrix elements involve a combination of classical sampling, Hamiltonian simulation, and the Hadamard test. These results have applications in developing a natural gradient descent algorithm for quantum Boltzmann machine learning, which takes into account the geometry of thermal states, and in establishing fundamental limitations on the ability to estimate the parameters of a Hamiltonian, when given access to thermal-state samples. For the latter task, and for the special case of estimating a single parameter, we sketch an algorithm that realizes a measurement that is asymptotically optimal for the estimation task. We finally stress that the natural gradient descent algorithm developed here can be used for any machine learning problem that employs the quantum Boltzmann machine ansatz.
Abstract Motivated by the growing interest in quantum machine learning, in particular quantum neural networks (QNNs), we study how recently introduced evaluation metrics based on the Fisher information matrix (FIM) are effective for predicting their training and prediction performance. We exploit the equivalence between a broad class of QNNs and Fourier models, and study the interplay between the effective dimension (ED) and the bias of a model towards a given task, investigating how these affect the model’s training and performance. We show that for a model that is completely agnostic, or unbiased, towards the function to be learned, a higher ED likely results in a better trainability and performance. On the other hand, for models that are biased towards the function to be learned a lower ED is likely beneficial during training. To obtain these results, we derive an analytical expression of the FIM for Fourier models and identify the features controlling a model’s ED. This allows us to construct models with tunable ED and bias, and to compare their training. We furthermore introduce a tensor network representation of the considered Fourier models, which could be a tool of independent interest for the analysis of QNN models. Overall, these findings provide an explicit example of the interplay between geometrical properties, model-task alignment and training, which are relevant for the broader machine learning community.
Energy-based models provide a natural bridge between statistical physics and machine learning by representing data through structured energy landscapes. Boltzmann machines are a particularly compelling class of such models for capturing complex interactions among latent variables, but their use in modern generative learning has been limited by the classical intractability of sampling from general (non-restricted) Boltzmann distributions. Here we develop a quantum-annealing-based framework that enables variational autoencoders with general Boltzmann priors. The framework employs three complementary annealing modes tailored to different stages of learning and deployment: diabatic quantum annealing provides unbiased Boltzmann samples for efficient training, slower annealing concentrates samples near low-energy configurations of the learned prior for unconditional generation, and conditional annealing with external fields steers the learned energy landscape toward attribute-specific regions for conditional generation and semantic editing. Using up to 2000 qubits on a D-Wave Advantage2 processor, we demonstrate stable training and high-quality generation on MNIST, Fashion-MNIST, and CelebA, achieving faster convergence and lower reconstruction loss than a Gaussian-prior VAE with the same encoder-decoder architecture. Beyond generation, the learned energy function provides out-of-distribution detection signals that add discriminative power beyond reconstruction loss. We demonstrate that these scores separate in-distribution samples from held-out digit classes in one-class MNIST experiments and improve the detection of market regime shifts in financial data. These results establish quantum annealing as a practical and controllable physical mechanism for energy-based representation learning and generative modeling beyond the reach of tractable classical approaches.
Abstract Quantum-inspired (QI) optimizers are increasingly proposed as alternatives to classical gradient-based training, yet are frequently benchmarked through classical stand-ins rather than themselves. Across tabular, vision, and language tasks we evaluate three classes: direct implementations of published update rules (Spall’s SPSA; quantum-behaved particle swarm optimization, QPSO, adapted to stochastic minibatches), a documented classical analog of quantum natural gradient (an NGD-style diagonal second-moment preconditioner), and the hyperparameter-control configurations under which such methods are often reported. The gap between these classes is our central result: on the tabular benchmark the hyperparameter control trains with the classical baselines while the direct algorithms fall ≈\approx ≈ 10–24 points below it, and on CIFAR-10/ResNet18, where v1’s “SPSA” arm was a learning-rate-scaled Adam, Adam itself stands ≈\approx ≈ 59 points above genuine SPSA. The QPSO adaptation stays at chance on LSTM tasks a legacy “QPSO” proxy solves at 79–81%, at 18–38 ×\times × Adam’s cost there and up to 74 ×\times × on the small CV model. A seed-paired 2×22\times 22×2 factorial on RoBERTa identifies the smaller learning rate as the dominant observed component of the earlier “QNG advantage” ( +0.50+0.50+0.50 points at the dataset level); with only three datasets, modified Hartung–Knapp uncertainty spans zero. The configured NGD arm shows no detectable advantage over a learning-rate-matched AdamW (descriptive pooled p=0.23p{=}0.23p=0.23 ) at 6– 7%7\%7% extra cost, and on ResNet18 its trained endpoints have larger dominant-curvature estimates than the Adam family’s. Headline comparisons use five seeds, paired tests, Holm correction, and stated detection limits. We conclude with an auditing taxonomy (direct algorithm / analog / hyperparameter control): within the arms, protocols, and scales evaluated, no advantage attributable to a QI mechanism is detected after these controls.
The Quantum Fisher Information matrix (QFIM) is a central metric in promising algorithms, such as Quantum Natural Gradient Descent and Variational Quantum Imaginary Time Evolution. Computing the full QFIM for a model with d parameters, however, is computationally expensive and generally requires O ( d 2 ) function evaluations. To remedy these increasing costs in high-dimensional parameter spaces, we propose using simultaneous perturbation stochastic approximation techniques to approximate the QFIM at a constant cost. We present the resulting algorithm and successfully apply it to prepare Hamiltonian ground states and train Variational Quantum Boltzmann Machines.