Pith. sign in

REVIEW 4 major objections 5 minor 54 references

EPIC-CIM: Training Convolutional Neural Networks on a Coherent Ising Machine via Equilibrium Propagation

T0 review · 4 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A convolutional neural network can be trained on a coherent Ising machine using equilibrium propagation, without backpropagation, and beats quantum baselines on MNIST, Fashion-MNIST, and CIFAR-10.

desk verdict The paper aims to train CNNs on a coherent Ising machine via equilibrium propagation, but the central update rule is unproven for most parameters and the comparisons are unfair; worth peer review, not citation. read the letter →

arxiv 2607.16271 v1 pith:XH4DERCD submitted 2026-07-08 quant-ph

Xingrui Yin , Shenwei Kang , Haoqi He , Yan Xiao , Hongdong Zhu , Hai Wei , Yin Ma , Qi Gao
show 2 more authors
Xiaochun Cao Kai Wen
This is my paper
classification quant-ph
keywords quantumconvolutionalneuralnetworksequilibriumpropagationcoherentIsingmachineenergy-basedlearningQUBOmodel
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

EPIC-CIM tries to establish that equilibrium propagation can train a convolutional neural network whose intermediate and output layers are sampled as binary states on a coherent Ising machine, eliminating the need for explicit gradient backpropagation. The network is defined by a single energy function that couples convolutional responses to binary latent states and output units under a one-hot constraint; training compares the free equilibrium with a weakly nudged equilibrium and updates parameters from the difference of spin-pair correlations. If correct, this gives a physically interpretable, hardware-native training rule for quantum CNNs that works in discrete, non-differentiable spaces, and the reported accuracies on MNIST, Fashion-MNIST, and CIFAR-10 beat established quantum baselines.

What carries the argument

The load-bearing object is the total energy function in Eq. 7, which combines convolutional feature matching, output coupling, and a one-hot constraint, together with the equilibrium propagation update in Eq. 12 that estimates parameter gradients from the difference between free and nudged equilibria. The paper assumes these two pieces connect: the energy's parameters are updated by spin-pair correlation differences obtained from CIM sampling.

What would settle it

Simulate a small convolutional problem where the true gradient of E_total with respect to k1 and W2 can be computed, then train with Eq. 12 and compare the resulting update direction against the true gradient; a significant mismatch would show the local rule does not implement EP for this energy.

Watch

Extended reading notes

Core claim

The central claim is that the full training of a QCNN—convolutional kernel, output weights, and biases—can be expressed as energy minimization and learned through equilibrium propagation. The paper introduces a total energy E_total = Conv(X1,k1)·X2 + (W2·flatten(X2)+b)·y + β(y^T1+(k-2))^2, and asserts that updating parameters by Δθ = (1/β)[(σ_iσ_j)_nudge − (σ_iσ_j)_free] drives learning. This replaces backprop with a local rule based on measurable state correlations. The evidence is experimental: accuracies of 92.32%, 79.62%, and 34.56% on the three datasets, with hardware experiments at 14-bit precision reaching 91% on a small MNIST subset.

Load-bearing premise

The paper assumes the equilibrium-propagation spin-correlation update (Eq. 12) applies to the parameters of the convolutional energy (Eq. 7), without a derivation showing that the energy derivatives equal differences of spin correlations, and that coherent Ising machine sampling reaches the two equilibrium states EP requires.

Editorial extensions

If this is right

  • Training quantum neural networks no longer requires differentiation; energy-based local rules suffice.
  • Full multi-class classification becomes feasible on CIM-style hardware.
  • The framework can extend to self-supervised and reinforcement learning, as the paper claims.
  • Sample efficiency improves: 88.64% on MNIST with only 2000 samples.
  • The gap to classical CNN narrows on CIFAR-10, suggesting viability for more complex visual tasks.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A testable consequence is that the EP rule should match exact gradients of E_total; computing ∂E/∂k1 directly and comparing with the correlation difference would confirm or refute the core mechanism.
  • The energy as written has no quadratic couplings among X2 units; if EP requires such couplings, the update may need additional terms—an extension the paper does not address.
  • The one-hot constraint with β might bias equilibria; tuning β could trade off classification margin against constraint satisfaction.
  • The method could generalize to deeper networks by stacking multiple energy layers with hidden-hidden couplings, following the EP formalism.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The manuscript proposes EPIC-CIM, a framework for training convolutional neural networks on a coherent Ising machine via equilibrium propagation. The energy function (Eq. 7) combines a convolutional feature-matching term, a linear output coupling term, and a one-hot output constraint. The paper claims that parameter updates can be obtained from CIM-measured spin correlations using the equilibrium-propagation rule (Eq. 12), without backpropagation. Experiments are reported on MNIST, Fashion-MNIST, CIFAR-10, including a hardware demonstration with 14-bit precision, with accuracy claims exceeding several quantum baselines.

Significance. If the central training mechanism were rigorously established, the paper would address an interesting and timely problem: training a physically implemented Ising machine with equilibrium propagation for convolutional tasks. The manuscript has the merit of proposing an explicit energy function and of reporting comparisons with several quantum baselines. However, the key derivation connecting Eq. (7) to Eq. (12) is absent, and the experimental protocol is not controlled enough to support the comparative accuracy claims. As written, the contribution is not established.

major comments (4)
  1. [§2.4, Eq. (12)] Equation (12) is asserted without derivation from Eq. (7). For the parameters in Eq. (7), ∂E_total/∂b_i = y_i and ∂E_total/∂k1 is a sum of products of input pixels and X2 spins; neither is a spin-pair correlation. Only ∂E_total/∂W2 is a product y_i·X2_j. Thus the proposed update rule cannot update b or k1, even if equilibrium propagation were otherwise valid. Moreover, Eq. (11) gives Δθ ∝ ∂E_free/∂θ − ∂E_nudge/∂θ, while Eq. (12) uses (nudge − free); the sign is inconsistent with the preceding gradient formula. In addition, the nudge term in Eq. (10) is scaled by ε, not by β, so the 1/β prefactor in Eq. (12) has no justification. These issues are load-bearing because the paper's central claim is that CIM-measured correlations implement EP training without backpropagation.
  2. [§2.4, Eqs. (9)–(10)] No argument is given that the CIM sampling realizes the two equilibrium states required by equilibrium propagation. The energy in Eq. (7) has no X2–X2 quadratic couplings; hidden variables interact only through the external field Conv(X1,k1) and the bilinear y–X2 term. The paper treats X2 and y as binary spin variables, but it does not show that the CIM steady state corresponds to sampling from the free or nudged energy, nor that differences of measured spin correlations estimate the gradient of any loss. The reference to Scellier and Bengio is not sufficient because the model does not satisfy the standard assumptions of EP.
  3. [§3.1, Table 1] The comparisons are confounded. The text states that the quantum baselines were trained on only a few thousand samples, while EPIC-CIM was trained on the full MNIST dataset; the training-set sizes for Fashion-MNIST and CIFAR-10 are not specified per baseline. A valid comparison requires identical data subsets, training/test splits, architectural capacities, and optimization budgets. Without these controls, the claimed improvements over D-Wave, QDEQs, and QCNN are not interpretable. The table also has an internal inconsistency: the note says QDEQs does not report training accuracy or standard deviation, yet the QDEQs column contains a training accuracy value.
  4. [§3.3, hardware validation] The real-hardware result (91% accuracy) is reported on 100 test samples without error bars, number of runs, confidence intervals, or device/calibration details. With only 100 test samples, a single run has a standard error of roughly 3 percentage points, so the claim that hardware performance is 'close to simulation' is not statistically supported. The '14-bit precision' statement is also not explained in terms of the CIM hardware's actual numerical or analog precision.
minor comments (5)
  1. [§2.3, Eq. (8)] The empirical risk in Eq. (8) is an MSE objective, but the paper does not explain how the energy in Eq. (7), whose free phase has no label-dependent term, is related to this loss. The connection between the energy model and the supervised objective should be made explicit.
  2. [§2.2, Eq. (5)] The dot products in Eq. (5) are used with different dimensional objects (Conv response with X2, and W2·flatten(X2)+b with y). Clarify the vectorization and dimension conventions.
  3. [Table 1] Several entries lack standard deviations, and the note about QDEQs training accuracy is inconsistent with the table body. Consider adding details on the number of runs and whether all entries come from the same protocol.
  4. [§3.1] Numerical precision: EPIC-CIM improves over D-Wave by 4.8 percentage points on MNIST, not 'approximately 5'; over QCNN by 7.7, not '7'. Use exact rounded values for consistency.
  5. [Throughout] The dataset name is consistently typeset as 'CIF AR-10'; it should be 'CIFAR-10'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reported accuracies are measured outcomes and the energy/update equations are not derived from the results they are used to explain.

full rationale

The paper's QUBO-to-Ising mapping (Eqs. 1–4) is standard algebraic transformation rather than a self-referential construction. The convolutional energy (Eq. 7) is explicitly constructed from input features, binary latent states, output couplings, and a one-hot regularizer; this is a model design choice, not a prediction derived from fitted outcomes. The equilibrium propagation update (Eq. 12) is adopted from the external EP literature (Scellier & Bengio, Laydevant et al.), not from the present authors' own prior results, and no fitting-to-prediction loop is present. The benchmark results in Table 1 and Fig. 3 are measured test accuracies, not quantities that are equivalent to the energy function or to the update rule by construction. The derivation has a genuine gap—Eq. 12 is asserted for all parameters while the derivatives of Eq. 7 with respect to k1 and b are not spin-pair correlations—but that is a correctness/derivation concern, not circularity: the accuracy numbers do not reduce to the assumed update rule. No load-bearing self-citations, ansatz-smuggling via self-citation, or renaming of a known result as a new derivation is evident.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central training mechanism rests on an unproven application of EP and on an unreported set of hyperparameters and hardware settings; the only standard content is the QUBO/Ising mapping.

free parameters (4)
  • β (output constraint strength)
    Introduced in Eq. 6 to enforce one-hot output; value never reported, yet it multiplies the constraint energy and appears in the denominator of the update rule Eq. 12.
  • ε (nudge strength)
    Introduced in Eq. 10 for the weakly clamped phase; no value or annealing schedule reported.
  • CNN architecture (kernel size, channels, strides, hidden spin count)
    Section 2.7/2.8 says real-valued convolutional layers are applied, but no architecture, number of spins/qubits, or layer counts are specified; performance depends on these choices.
  • CIM parameters (precision, repetitions, coupling strength)
    Section 3.3 mentions 14-bit precision and 1000/100 sample split, but no device parameters or run counts are given.
assumptions (4)
  • standard math QUBO-to-Ising variable mapping x_i = (1+s_i)/2 and coefficient transformations Eq. 2–4
    Standard textbook map, used without proof in §2.1.
  • domain assumption Equilibrium propagation update Δθ = (1/β)[(σ_iσ_j)_nudge − (σ_iσ_j)_free] for all network parameters
    Stated in §2.4 Eq. 12 without derivation; does not follow from Eq. 7 because the energy lacks hidden-hidden quadratic couplings, so it is unclear which spin products correspond to k1 and W2.
  • domain assumption CIM quantum sampling produces the free and nudged equilibrium distributions required by EP
    Section 2.7 treats CIM outputs as samples from the low-energy distribution; EP requires deterministic fixed-point equilibria of the same energy function, and no evidence connects CIM ground-state optimization to these equilibria.
  • ad hoc to paper One-hot constraint β(y^T 1 + (k−2))^2 yields a valid k-class classifier
    Imposed in Eq. 6; used to construct the energy but its sufficiency for multiclass classification is not analyzed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of EPIC-CIM: Training Convolutional Neural Networks on a Coherent Ising Machine via Equilibrium Propagation." pith.science (2026). https://pith.science/paper/XH4DERCD

@misc{pith2026260716271,
  author       = {Pith},
  title        = {Pith review of: EPIC-CIM: Training Convolutional Neural Networks on a Coherent Ising Machine via Equilibrium Propagation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XH4DERCD}},
  note         = {Machine review of arXiv:2607.16271}
}
read the original abstract

Quantum convolutional neural networks, due to the involvement of quantum measurements and discrete quantum state evolution, face inherent training challenges associated with non-differentiable operations and discrete optimization dynamics, which make conventional gradient-based learning difficult to apply effectively. In this context, energy-based learning provides a promising alternative by reformulating network training as an energy minimization process without explicit gradient backpropagation.In this framework, input data are processed through convolutional operations, followed by quantum sampling to generate intermediate binary representations, while the output layer also relies on quantum sampling to produce final predictions. The overall network energy is composed of convolutional feature matching terms, linear coupling terms at the output layer, and global output constraint terms, enabling both parameter updates and feature evolution to be described through physically interpretable energy dynamics. Furthermore, under the equilibrium propagation mechanism, the energy difference between the free phase and the weakly clamped phase is exploited to drive parameter updates without explicit gradient computation, thereby enabling stable and consistent learning in non-differentiable and discrete spaces. While remaining consistent with classical convolutional learning theory, the proposed framework enhances interpretability and observability through quantum energy modeling, offering a unified physical perspective for efficient QCNN training and the integration of quantum computing with artificial intelligence.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

54 extracted references · 3 linked inside Pith

  1. [1]

    Proceedings of the IEEE109(3), 247–278 (2021)

    Samek, W., Montavon, G., Lapuschkin, S., Anders, C.J., M¨ uller, K.-R.: Explain- ing deep neural networks and beyond: A review of methods and applications. Proceedings of the IEEE109(3), 247–278 (2021)

    2021

  2. [2]

    Fundamentals of artificial intelli- gence, 603–649 (2020) 15

    Chowdhary, K.: Natural language processing. Fundamentals of artificial intelli- gence, 603–649 (2020) 15

    2020

  3. [3]

    Nature Reviews Neuroscience21(6), 335–346 (2020)

    Lillicrap, T.P., Santoro, A., Marris, L., Akerman, C.J., Hinton, G.: Backpropa- gation and the brain. Nature Reviews Neuroscience21(6), 335–346 (2020)

    2020

  4. [4]

    Advances in neural information processing systems29(2016)

    Hubara, I., Courbariaux, M., Soudry, D., El-Yaniv, R., Bengio, Y.: Binarized neural networks. Advances in neural information processing systems29(2016)

    2016

  5. [6]

    Nature Communications15(1), 3671 (2024)

    Laydevant, J., Markovi´ c, D., Grollier, J.: Training an ising machine with equilibrium propagation. Nature Communications15(1), 3671 (2024)

    2024

  6. [7]

    Frontiers in computational neuro- science11, 24 (2017)

    Scellier, B., Bengio, Y.: Equilibrium propagation: Bridging the gap between energy-based models and backpropagation. Frontiers in computational neuro- science11, 24 (2017)

    2017

  7. [8]

    Structural optimization8(2), 69–85 (1994)

    Arora, J., Huang, M., Hsieh, C.: Methods for optimization of nonlinear problems with discrete variables: a review. Structural optimization8(2), 69–85 (1994)

    1994

  8. [9]

    Information Sciences236, 126–137 (2013)

    Valdez, S.I., Hern´ andez, A., Botello, S.: A boltzmann based estimation of distribution algorithm. Information Sciences236, 126–137 (2013)

    2013

Show all 54 references
  1. [10]

    In: Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp

    Kappes, J., Andres, B., Hamprecht, F., Schnorr, C., Nowozin, S., Batra, D., Kim, S., Kausler, B., Lellmann, J., Komodakis, N.,et al.: A comparative study of modern inference techniques for discrete energy minimization problems. In: Pro- ceedings of the IEEE Conference on Compu...

    2013

  2. [11]

    Scientific reports11(1), 10029 (2021)

    Date, P., Arthur, D., Pusey-Nazzaro, L.: Qubo formulations for training machine learning models. Scientific reports11(1), 10029 (2021)

    2021

  3. [12]

    arXiv preprint arXiv:1811.11538 (2018)

    Glover, F., Kochenberger, G., Du, Y.: A tutorial on formulating and using qubo models. arXiv preprint arXiv:1811.11538 (2018)

    2018 arXiv

  4. [13]

    Physics Reports1068, 1–66 (2024)

    Blekos, K., Brand, D., Ceschini, A., Chou, C.-H., Li, R.-H., Pandya, K., Summer, A.: A review on quantum approximate optimization algorithm and its variants. Physics Reports1068, 1–66 (2024)

    2024

  5. [14]

    npj Quantum Information8(1), 62 (2022)

    Du, Y., Huang, T., You, S., Hsieh, M.-H., Tao, D.: Quantum circuit architecture search for variational quantum algorithms. npj Quantum Information8(1), 62 (2022)

    2022

  6. [15]

    Yann, L.: MNIST handwritten digit database (2010)

    2010

  7. [16]

    Reviews of Modern Physics94(1), 015004 (2022) 16

    Bharti, K., Cervera-Lierta, A., Kyaw, T.H., Haug, T., Alperin-Lea, S., Anand, A., Degroote, M., Heimonen, H., Kottmann, J.S., Menke, T.,et al.: Noisy intermediate-scale quantum algorithms. Reviews of Modern Physics94(1), 015004 (2022) 16

    2022

  8. [17]

    Nature communications 9(1), 4812 (2018)

    McClean, J.R., Boixo, S., Smelyanskiy, V.N., Babbush, R., Neven, H.: Barren plateaus in quantum neural network training landscapes. Nature communications 9(1), 4812 (2018)

    2018

  9. [18]

    Optics Communications550, 129993 (2024)

    Gong, L.-H., Pei, J.-J., Zhang, T.-F., Zhou, N.-R.: Quantum convolutional neu- ral network based on variational quantum circuits. Optics Communications550, 129993 (2024)

    2024

  10. [19]

    IEEE access8, 141007–141024 (2020)

    Chen, S.Y.-C., Yang, C.-H.H., Qi, J., Chen, P.-Y., Ma, X., Goan, H.-S.: Variational quantum circuits for deep reinforcement learning. IEEE access8, 141007–141024 (2020)

    2020

  11. [20]

    Quantum Information Processing23(5), 189 (2024)

    Shi, S., Wang, Z., Li, J., Li, Y., Shang, R., Zhong, G., Gu, Y.: Quantum convolu- tional neural networks for multiclass image classification. Quantum Information Processing23(5), 189 (2024)

    2024

  12. [21]

    Fracture and Structural Integrity17(64), 51–76 (2023)

    Ghannadi, P., Kourehli, S.S., Mirjalili, S.: A review of the application of the sim- ulated annealing algorithm in structural health monitoring (1995-2021). Fracture and Structural Integrity17(64), 51–76 (2023)

    1995

  13. [22]

    Philosophical Transactions of the Royal Society A381(2241), 20210417 (2023)

    Rajak, A., Suzuki, S., Dutta, A., Chakrabarti, B.K.: Quantum annealing: An overview. Philosophical Transactions of the Royal Society A381(2241), 20210417 (2023)

    2023

  14. [23]

    arXiv preprint arXiv:2311.03408 (2023)

    Song, X., Liu, T., Li, S.E., Duan, J., Wang, W., Li, K.: Training multi-layer neural networks on ising machine. arXiv preprint arXiv:2311.03408 (2023)

    2023 arXiv

  15. [24]

    Computational optimization and applications47(2), 257–288 (2010)

    Murray, W., Ng, K.-M.: An algorithm for nonlinear optimization problems with binary variables. Computational optimization and applications47(2), 257–288 (2010)

    2010

  16. [25]

    IEEE transactions on neural networks and learning systems33(12), 6999–7019 (2021)

    Li, Z., Liu, F., Yang, W., Peng, S., Zhou, J.: A survey of convolutional neu- ral networks: analysis, applications, and prospects. IEEE transactions on neural networks and learning systems33(12), 6999–7019 (2021)

    2021

  17. [26]

    Frontiers in Physics13, 1529188 (2025)

    Lee, C., Araujo, I.F., Kim, D., Lee, J., Park, S., Ryu, J.-Y., Park, D.K.: Opti- mizing quantum convolutional neural network architectures for arbitrary data dimension. Frontiers in Physics13, 1529188 (2025)

    2025

  18. [27]

    Quantum Machine Intelligence4(1), 3 (2022)

    Hur, T., Kim, L., Park, D.K.: Quantum convolutional neural network for classical data classification. Quantum Machine Intelligence4(1), 3 (2022)

    2022

  19. [28]

    Science advances7(40), 0952 (2021)

    Honjo, T., Sonobe, T., Inaba, K., Inagaki, T., Ikuta, T., Yamada, Y., Kazama, T., Enbutsu, K., Umeki, T., Kasahara, R.,et al.: 100,000-spin coherent ising machine. Science advances7(40), 0952 (2021)

    2021

  20. [29]

    Nature Reviews Physics4(6), 363–379 (2022)

    Mohseni, N., McMahon, P.L., Byrnes, T.: Ising machines as hardware solvers 17 of combinatorial optimization problems. Nature Reviews Physics4(6), 363–379 (2022)

    2022

  21. [30]

    Light: Science & Applications11(1), 333 (2022)

    Cen, Q., Ding, H., Hao, T., Guan, S., Qin, Z., Lyu, J., Li, W., Zhu, N., Xu, K., Dai, Y.,et al.: Large-scale coherent ising machine based on optoelectronic parametric oscillator. Light: Science & Applications11(1), 333 (2022)

    2022

  22. [31]

    npj Complexity1(1), 10 (2024)

    Macy, M.W., Szymanski, B.K., Ho lyst, J.A.: The ising model celebrates a century of interdisciplinary contributions. npj Complexity1(1), 10 (2024)

    2024

  23. [32]

    Physical Review Letters134(11), 117103 (2025)

    Avni, Y., Fruchart, M., Martin, D., Seara, D., Vitelli, V.: Nonreciprocal ising model. Physical Review Letters134(11), 117103 (2025)

    2025

  24. [33]

    Punnen, A.P.: The Quadratic Unconstrained Binary Optimization Problem vol. 10, pp. 978–3. Springer, ??? (2022)

    2022

  25. [34]

    Bollettino dell’Unione Matematica Italiana18(2), 615–635 (2025)

    Isopi, M., Scoppola, B., Troiani, A.: On some features of quadratic unconstrained binary optimization with random coefficients. Bollettino dell’Unione Matematica Italiana18(2), 615–635 (2025)

    2025

  26. [35]

    IEEE access11, 104165–104178 (2023)

    Jiang, J.-R., Chu, C.-W.: Classifying and benchmarking quantum annealing algo- rithms based on quadratic unconstrained binary optimization for solving np-hard problems. IEEE access11, 104165–104178 (2023)

    2023

  27. [36]

    Frontiers in physics2, 5 (2014)

    Lucas, A.: Ising formulations of many np problems. Frontiers in physics2, 5 (2014)

    2014

  28. [37]

    npj Quantum Information3(1), 15 (2017)

    Lund, A.P., Bremner, M.J., Ralph, T.C.: Quantum sampling problems, boson- sampling and quantum supremacy. npj Quantum Information3(1), 15 (2017)

    2017

  29. [38]

    Physical Review A—Atomic, Molecular, and Optical Physics78(4), 042336 (2008)

    Wocjan, P., Abeyesinghe, A.: Speedup via quantum sampling. Physical Review A—Atomic, Molecular, and Optical Physics78(4), 042336 (2008)

    2008

  30. [39]

    SIAM journal on optimization11(1), 179–188 (2000)

    Beck, A., Teboulle, M.: Global optimality conditions for quadratic optimization problems with binary constraints. SIAM journal on optimization11(1), 179–188 (2000)

    2000

  31. [40]

    INFORMS Journal on Computing33(1), 401–418 (2021)

    Bergman, D., Lozano, L.: Decision diagram decomposition for quadratically con- strained binary optimization. INFORMS Journal on Computing33(1), 401–418 (2021)

    2021

  32. [41]

    Nature Reviews Physics, 1–18 (2024)

    Abbas, A., Ambainis, A., Augustino, B., B¨ artschi, A., Buhrman, H., Coffrin, C., Cortiana, G., Dunjko, V., Egger, D.J., Elmegreen, B.G., et al.: Challenges and opportunities in quantum optimization. Nature Reviews Physics, 1–18 (2024)

    2024

  33. [42]

    Ieee Access8, 23568–23593 (2020) 18

    Li, Y., Tian, M., Liu, G., Peng, C., Jiao, L.: Quantum optimization and quantum learning: A survey. Ieee Access8, 23568–23593 (2020) 18

    2020

  34. [43]

    Geoscientific Model Development Discussions2022, 1–10 (2022)

    Hodson, T.O.: Root mean square error (rmse) or mean absolute error (mae): When to use them or not. Geoscientific Model Development Discussions2022, 1–10 (2022)

    2022

  35. [44]

    Li, S.E.: Reinforcement learning for sequential decision and optimal control (2023)

    2023

  36. [45]

    International Journal of Intelligent Systems36(8), 4439–4467 (2021)

    Guan, Y., Li, S.E., Duan, J., Li, J., Ren, Y., Sun, Q., Cheng, B.: Direct and indi- rect reinforcement learning. International Journal of Intelligent Systems36(8), 4439–4467 (2021)

    2021

  37. [46]

    In: Machine Learn- ing Techniques for Multimedia: Case Studies on Organization and Retrieval, pp

    Cunningham, P., Cord, M., Delany, S.J.: Supervised learning. In: Machine Learn- ing Techniques for Multimedia: Case Studies on Organization and Retrieval, pp. 21–49. Springer, ??? (2008)

    2008

  38. [47]

    Horizons

    Nasteski, V.: An overview of the supervised machine learning methods. Horizons. b4(51-62), 56 (2017)

    2017

  39. [48]

    arXiv preprint arXiv:2307.01668 (2023)

    Luo, W., Jiang, H., Hu, T., Sun, J., Li, Z., Zhang, Z.: Training energy-based models with diffusion contrastive divergences. arXiv preprint arXiv:2307.01668 (2023)

    2023 arXiv

  40. [49]

    In: International Workshop on Artificial Intelligence and Statistics, pp

    Carreira-Perpinan, M.A., Hinton, G.: On contrastive divergence learning. In: International Workshop on Artificial Intelligence and Statistics, pp. 33–40 (2005). PMLR

    2005

  41. [50]

    International Journal of Education & Language Studies1(2), 78–82 (2022)

    Noor, S., Tajik, O., Golzar, J.: Simple random sampling. International Journal of Education & Language Studies1(2), 78–82 (2022)

    2022

  42. [51]

    Current Directions in Psychological Science15(5), 207–211 (2006)

    Spivey, M.J., Dale, R.: Continuous dynamics in real-time cognition. Current Directions in Psychological Science15(5), 207–211 (2006)

    2006

  43. [52]

    Sensors22(23), 9544 (2022)

    Nocentini, O., Kim, J., Bashir, M.Z., Cavallo, F.: Image classification using mul- tiple convolutional neural networks on the fashion-mnist dataset. Sensors22(23), 9544 (2022)

    2022

  44. [53]

    Unpublished manuscript40(7), 1–9 (2010)

    Krizhevsky, A., Hinton, G.,et al.: Convolutional deep belief networks on cifar-10. Unpublished manuscript40(7), 1–9 (2010)

    2010

  45. [54]

    Advances in Neural Information Processing Systems37, 31940–31967 (2024)

    Schleich, P., Skreta, M., Kristensen, L., Vargas-Hernandez, R., Aspuru-Guzik, A.: Quantum deep equilibrium models. Advances in Neural Information Processing Systems37, 31940–31967 (2024)

    2024

  46. [55]

    Quantum Science and Technology10(2), 025050 (2025) 19

    Monbroussou, L., Landman, J., Wang, L., Grilo, A.B., Kashefi, E.: Subspace preserving quantum convolutional neural network architectures. Quantum Science and Technology10(2), 025050 (2025) 19

    2025

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.