pith. sign in

arxiv: 2506.19461 · v4 · submitted 2025-06-24 · 🪐 quant-ph · cs.AI· stat.ML

Iterative Quantum Feature Maps

Nasa Matsumoto , Quoc Hoan Tran , Koki Chinzei , Yasuhiro Endo , Hirotaka Oshima This is my paper

Pith reviewed 2026-05-19 08:13 UTC · model grok-4.3

classification 🪐 quant-ph cs.AIstat.ML
keywords quantum machine learningquantum feature mapshybrid quantum-classicalcontrastive learningnoisy intermediate-scale quantumimage classificationlayer-wise training
0
0 comments X

The pith

Iterative Quantum Feature Maps build deep quantum models by linking shallow circuits with classical weights and layer-wise training.

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

The paper introduces Iterative Quantum Feature Maps as a hybrid framework that stacks shallow quantum feature maps step by step, using weights computed on a classical computer. Contrastive learning and training one layer at a time let the method cut quantum runtime and limit how much hardware noise hurts the results. Experiments on noisy quantum data show the approach beats quantum convolutional neural networks while skipping any optimization of variational parameters. On ordinary image classification benchmarks, a tuned version reaches accuracy levels similar to standard classical neural networks.

Core claim

Iterative Quantum Feature Maps (IQFMs) constructs a deep architecture by iteratively connecting shallow quantum feature maps with classically computed augmentation weights. Incorporating contrastive learning and layer-wise training reduces quantum runtime and mitigates noise-induced degradation. Numerical experiments on noisy quantum data tasks show IQFMs outperforms quantum convolutional neural networks without optimizing variational quantum parameters. On classical image classification benchmarks, carefully designed IQFMs achieve performance comparable to classical neural networks.

What carries the argument

Iteratively connecting shallow quantum feature maps using classically computed augmentation weights, supported by contrastive learning and layer-wise training.

If this is right

  • Only shallow circuits run at each iteration, lowering total quantum runtime.
  • Noise effects decrease because deep circuits are avoided.
  • No variational quantum parameters are optimized, removing a major training bottleneck.
  • Performance on noisy quantum tasks exceeds quantum convolutional neural networks.
  • Accuracy on standard image benchmarks reaches levels comparable to classical neural networks.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The method could let current noisy hardware support deeper quantum models without building actual deep circuits.
  • Hybrid classical augmentation of quantum layers might extend to regression or generative tasks beyond classification.
  • Layer-wise training could lower overall quantum resource costs enough to make near-term quantum machine learning more feasible.

Load-bearing premise

That repeatedly linking simple quantum circuits through classical weights plus contrastive and staged training can keep useful quantum behavior while overcoming hardware noise.

What would settle it

A side-by-side test on the same noisy quantum classification dataset in which IQFMs fails to exceed the accuracy of a quantum convolutional neural network or shows no reduction in quantum resources used.

Figures

Figures reproduced from arXiv: 2506.19461 by Hirotaka Oshima, Koki Chinzei, Nasa Matsumoto, Quoc Hoan Tran, Yasuhiro Endo.

Figure 1
Figure 1. Figure 1: FIG. 1: IQFMs and the representation learning in processing classical input [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2: The modular IQFMs. The classical data [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3: The quantum feature extraction involves [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4: Ground-state phase diagrams of Tasks A and B. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5: For quantum data classification tasks, IQFMs incorporates a re-input structure: [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6: Test accuracy for Tasks A and B using IQFMs [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8: Test accuracy for (a) Task A and (b) Task B of [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7: Visualization of the final representation vectors [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10: Bar plot comparing the average test accuracy [PITH_FULL_IMAGE:figures/full_fig_p009_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIG. 11: Test accuracy for (a) Task A and (b) Task B [PITH_FULL_IMAGE:figures/full_fig_p011_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: FIG. 12: Test accuracy for (a) Task A and (b) Task B [PITH_FULL_IMAGE:figures/full_fig_p011_12.png] view at source ↗
read the original abstract

Quantum machine learning models that leverage quantum circuits as quantum feature maps (QFMs) are recognized for their enhanced expressive power in learning tasks. Such models have demonstrated rigorous end-to-end quantum speedups for specific families of classification problems. However, deploying deep QFMs on real quantum hardware remains challenging due to circuit noise and hardware constraints. Additionally, variational quantum algorithms often suffer from computational bottlenecks, particularly in accurate gradient estimation, which significantly increases quantum resource demands during training. We propose Iterative Quantum Feature Maps (IQFMs), a hybrid quantum-classical framework that constructs a deep architecture by iteratively connecting shallow QFMs with classically computed augmentation weights. By incorporating contrastive learning and a layer-wise training mechanism, the IQFMs framework effectively reduces quantum runtime and mitigates noise-induced degradation. In tasks involving noisy quantum data, numerical experiments show that the IQFMs framework outperforms quantum convolutional neural networks, without requiring the optimization of variational quantum parameters. Even for a typical classical image classification benchmark, a carefully designed IQFMs framework achieves performance comparable to that of classical neural networks. This framework presents a promising path to address current limitations and harness the full potential of quantum-enhanced machine learning.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes Iterative Quantum Feature Maps (IQFMs), a hybrid quantum-classical framework that constructs deep architectures by iteratively connecting shallow quantum feature maps (QFMs) using classically computed augmentation weights, together with contrastive learning and layer-wise training. The central claims are that this approach reduces quantum runtime, mitigates noise degradation, outperforms quantum convolutional neural networks on noisy quantum data without optimizing variational parameters, and achieves performance comparable to classical neural networks on standard image classification benchmarks.

Significance. If the empirical results hold and the iterative hybrid construction demonstrably retains quantum expressivity beyond classical post-processing of shallow kernels, the framework could provide a practical route to scaling quantum machine learning on noisy intermediate-scale hardware by sidestepping deep-circuit noise and variational training overheads. The combination of layer-wise training and contrastive learning is a potentially useful design choice for hybrid models.

major comments (3)
  1. [Abstract and §5] Abstract and §5 (Numerical Experiments): performance claims rest on unspecified numerical experiments with no reported details on dataset sizes, classical baselines, error bars, statistical tests, or controls for the classical augmentation and contrastive components. This prevents assessment of whether the reported outperformance over QCNNs is supported by the data or attributable to the quantum feature maps.
  2. [§3] §3 (IQFMs Framework): each iteration requires measurement after the shallow QFM to extract features for classical augmentation weights. This projects the quantum state and breaks coherence between layers, so the overall map is equivalent to a classical composition of independent shallow quantum kernels rather than a single coherent deep quantum feature map. The paper must show either that inter-layer quantum interference is preserved or that the effective expressivity exceeds the sum of the shallow circuits; otherwise the noise-mitigation and outperformance claims cannot be attributed to the quantum component.
  3. [§4] §4 (Contrastive Learning and Layer-wise Training): the assumption that classically computed augmentation weights combined with contrastive loss can construct an effective deep architecture while retaining useful quantum expressivity is load-bearing for both the runtime reduction and the performance gains, yet no analysis or ablation is provided to separate the contribution of the quantum kernels from the classical contrastive training.
minor comments (2)
  1. [§3] Notation for the augmentation weights and the iterative update rule should be defined explicitly with equations rather than prose descriptions to improve reproducibility.
  2. [Figure 2] Figure captions for the architecture diagrams should include the circuit depth and number of qubits used in each shallow QFM block.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed comments on our manuscript. We have carefully addressed each major point below, providing clarifications and indicating where revisions have been made to strengthen the presentation and analysis.

read point-by-point responses

  1. Referee: [Abstract and §5] Abstract and §5 (Numerical Experiments): performance claims rest on unspecified numerical experiments with no reported details on dataset sizes, classical baselines, error bars, statistical tests, or controls for the classical augmentation and contrastive components. This prevents assessment of whether the reported outperformance over QCNNs is supported by the data or attributable to the quantum feature maps.

    Authors: We agree that the original manuscript lacked sufficient experimental details. In the revised version, we have substantially expanded §5 and added a dedicated appendix with full specifications: dataset sizes and preprocessing steps, descriptions of all classical baselines (including standard CNNs, kernel SVMs, and random feature models), error bars computed over 10–20 independent runs with standard deviations, results of statistical significance tests (e.g., paired t-tests), and explicit control experiments that disable either the quantum feature maps or the contrastive component while keeping the rest of the pipeline fixed. These additions allow direct assessment of the quantum contribution. revision: yes

  2. Referee: [§3] §3 (IQFMs Framework): each iteration requires measurement after the shallow QFM to extract features for classical augmentation weights. This projects the quantum state and breaks coherence between layers, so the overall map is equivalent to a classical composition of independent shallow quantum kernels rather than a single coherent deep quantum feature map. The paper must show either that inter-layer quantum interference is preserved or that the effective expressivity exceeds the sum of the shallow circuits; otherwise the noise-mitigation and outperformance claims cannot be attributed to the quantum component.

    Authors: We acknowledge that intermediate measurements collapse the quantum state and therefore preclude coherent interference between successive layers. The IQFM construction is intentionally hybrid: each shallow QFM extracts quantum-enhanced features that are then combined via classically computed augmentation weights. We argue that the effective expressivity still exceeds a purely classical post-processing of the same shallow kernels because the contrastive layer-wise training optimizes the augmentation weights to exploit the specific geometry induced by each quantum feature map. In the revised §3 we have added a short theoretical paragraph and a supporting lemma showing that the composed map can realize functions outside the span of independent shallow-kernel compositions when the augmentation weights are learned under the contrastive objective. We therefore maintain that the observed noise robustness and performance gains can be attributed to the quantum components, even without inter-layer coherence. revision: partial

  3. Referee: [§4] §4 (Contrastive Learning and Layer-wise Training): the assumption that classically computed augmentation weights combined with contrastive loss can construct an effective deep architecture while retaining useful quantum expressivity is load-bearing for both the runtime reduction and the performance gains, yet no analysis or ablation is provided to separate the contribution of the quantum kernels from the classical contrastive training.

    Authors: We agree that ablation studies are required to substantiate the load-bearing assumption. The revised manuscript now includes a new subsection in §4 together with an appendix that reports systematic ablations: (i) IQFM with quantum feature maps replaced by classical random features, (ii) removal of the contrastive loss while retaining the iterative augmentation, and (iii) layer-wise training versus end-to-end training. The results show that both the quantum kernels and the contrastive objective contribute measurably to the final accuracy, with the largest drop occurring when the quantum feature maps are removed. These experiments directly separate the contributions and support the runtime-reduction claims. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical framework validated by external numerical benchmarks

full rationale

The paper proposes the IQFMs hybrid framework as an iterative construction of shallow QFMs connected via classically computed augmentation weights, with contrastive learning and layer-wise training. Its central claims are framed as outcomes of numerical experiments on noisy quantum data (outperforming QCNNs without variational parameter optimization) and classical image classification benchmarks (comparable to classical NNs). No load-bearing derivation, equation, or prediction reduces by the paper's own definitions to a fitted parameter or self-referential input; the architecture is a design choice whose performance is assessed against independent external benchmarks rather than constructed tautologically. Self-citations, if present, are not invoked as uniqueness theorems or load-bearing justifications for the core results.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The central claim rests on the empirical effectiveness of the hybrid iterative construction rather than on new axioms or derivations; no free parameters, background axioms, or invented entities are explicitly introduced in the abstract.

pith-pipeline@v0.9.0 · 5744 in / 1294 out tokens · 42525 ms · 2026-05-19T08:13:18.982269+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Recurrent Quantum Feature Maps for Reservoir Computing

    quant-ph 2026-04 unverdicted novelty 6.0

    Recurrent quantum feature maps achieve lower mean squared error than echo state networks and multilayer perceptrons on Mackey-Glass prediction using compact quantum circuits.

Reference graph

Works this paper leans on

53 extracted references · 53 canonical work pages · cited by 1 Pith paper · 5 internal anchors

  1. [1]

    At the layer l = 1, the QFM is initialized with a quantum data |ψ⟩0 and a classical data h0

    work page

  2. [2]

    , L, the following operations are performed: gl = fl(|ψ⟩l−1 , hl−1, Pl, θl) ∈ Rdg , (1) hl = Al(Wl, gl) ∈ Rdh

    For each layer l = 1, . . . , L, the following operations are performed: gl = fl(|ψ⟩l−1 , hl−1, Pl, θl) ∈ Rdg , (1) hl = Al(Wl, gl) ∈ Rdh . (2) Here, Eqs. (1) and (2) represent the quantum feature ex- traction via the QFM of the lth layer and the classical augmentation Al with parameters Wl (such as a clas- sical neural network), respectively. Additionall...

    work page

  3. [3]

    Havl´ ıˇ cek, A

    V. Havl´ ıˇ cek, A. D. C´ orcoles, K. Temme, A. W. Harrow, A. Kandala, J. M. Chow, and J. M. Gambetta, Super- vised learning with quantum-enhanced feature spaces, Nature 567, 209 (2019)

    work page 2019

  4. [4]

    Schuld and N

    M. Schuld and N. Killoran, Quantum machine learning in feature Hilbert spaces, Phys. Rev. Lett. 122, 040504 (2019)

    work page 2019

  5. [5]

    Bravyi, D

    S. Bravyi, D. Gosset, and R. K¨ onig, Quantum advantage with shallow circuits, Science 362, 308–311 (2018)

    work page 2018

  6. [6]

    Y. Liu, S. Arunachalam, and K. Temme, A rigorous and robust quantum speed-up in supervised machine learn- ing, Nat. Phys. (2021)

    work page 2021

  7. [7]

    T. Goto, Q. H. Tran, and K. Nakajima, Universal approx- imation property of quantum machine learning models in quantum-enhanced feature spaces, Phys. Rev. Lett. 127, 090506 (2021)

    work page 2021

  8. [8]

    Huang, M

    H.-Y. Huang, M. Broughton, M. Mohseni, R. Babbush, S. Boixo, H. Neven, and J. R. McClean, Power of data in quantum machine learning, Nat. Commun. 12, 2631 (2021)

    work page 2021

  9. [9]

    Mitarai, M

    K. Mitarai, M. Negoro, M. Kitagawa, and K. Fujii, Quan- tum circuit learning, Phys. Rev. A 98, 032309 (2018)

    work page 2018

  10. [10]

    E. R. Anschuetz and B. T. Kiani, Quantum variational algorithms are swamped with traps, Nat. Commun. 13, 7760 (2022)

    work page 2022

  11. [11]

    Bittel and M

    L. Bittel and M. Kliesch, Training variational quantum algorithms is np-hard, Phys. Rev. Lett. 127, 120502 (2021)

    work page 2021

  12. [12]

    J. R. McClean, S. Boixo, V. N. Smelyanskiy, R. Babbush, and H. Neven, Barren plateaus in quantum neural net- work training landscapes, Nat. Commun. 9, 4812 (2018)

    work page 2018

  13. [13]

    Cerezo, A

    M. Cerezo, A. Sone, T. Volkoff, L. Cincio, and P. J. Coles, Cost function dependent barren plateaus in shal- low parametrized quantum circuits, Nat. Commun. 12, 1791 (2021)

    work page 2021

  14. [14]

    Ortiz Marrero, M

    C. Ortiz Marrero, M. Kieferov´ a, and N. Wiebe, Entanglement-induced barren plateaus, PRX Quantum 2, 040316 (2021)

    work page 2021

  15. [15]

    Gil-Fuster, C

    E. Gil-Fuster, C. Gyurik, A. P´ erez-Salinas, and V. Dun- jko, On the relation between trainability and dequan- tization of variational quantum learning models, arXiv 10.48550/arXiv.2406.07072 (2024)

    work page doi:10.48550/arxiv.2406.07072 2024

  16. [16]

    Quantum Convolutional Neural Networks are Effectively Classically Simulable

    P. Bermejo, P. Braccia, M. S. Rudolph, Z. Holmes, L. Cincio, and M. Cerezo, Quantum convolutional neu- ral networks are (effectively) classically simulable, arXiv 10.48550/arXiv.2408.12739 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2408.12739 2024

  17. [17]

    Lerch, R

    S. Lerch, R. Puig, M. S. Rudolph, A. Angrisani, T. Jones, M. Cerezo, S. Thanasilp, and Z. Holmes, Efficient quantum-enhanced classical simulation for patches of quantum landscapes, arXiv 10.48550/arXiv.2411.19896 (2024)

    work page doi:10.48550/arxiv.2411.19896 2024

  18. [18]

    Abbas, R

    A. Abbas, R. King, H.-Y. Huang, W. J. Hug- gins, R. Movassagh, D. Gilboa, and J. Mc- Clean, On quantum backpropagation, informa- tion reuse, and cheating measurement collapse, in Advances in Neural Information Processing Systems, Vol. 36 (2023)

    work page 2023

  19. [19]

    Chinzei, S

    K. Chinzei, S. Yamano, Q. H. Tran, Y. Endo, and H. Os- hima, Trade-off between gradient measurement efficiency and expressivity in deep quantum neural networks, npj Quantum Inf. 11, 79 (2025)

    work page 2025

  20. [20]

    Liu, Z.-Y

    H.-Y. Liu, Z.-Y. Chen, T.-P. Sun, C. Xue, Y.-C. Wu, and G.-P. Guo, Can variational quantum algorithms demon- strate quantum advantages? time really matters, arXiv 10.48550/arxiv.2307.04089 (2023)

    work page doi:10.48550/arxiv.2307.04089 2023

  21. [21]

    Hadsell, S

    R. Hadsell, S. Chopra, and Y. LeCun, Dimension- ality reduction by learning an invariant mapping, in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Vol. 2 (IEEE, 2006) pp. 1735–1742

    work page 2006

  22. [22]

    Gutmann and A

    M. Gutmann and A. Hyv¨ arinen, Noise- contrastive estimation: A new estimation prin- ciple for unnormalized statistical models, in Proc. 13th Int. Conf. Artif. Intell. Stat., Vol. 9 (PMLR,

    work page

  23. [23]

    Representation Learning with Contrastive Predictive Coding

    A. van den Oord, Y. Li, and O. Vinyals, Representa- tion learning with contrastive predictive coding, arXiv 10.48550/arXiv.1807.03748 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1807.03748 2018

  24. [24]

    K. He, H. Fan, Y. Wu, S. Xie, and R. Girshick, Momentum contrast for unsu- pervised visual representation learning, in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR) (IEEE, 2020) pp. 9729–9738

    work page 2020

  25. [25]

    T. Chen, S. Kornblith, M. Norouzi, and G. Hinton, A simple framework for con- trastive learning of visual representations, in Proc. 37th Int. Conf. Mach. Learn. (ICML), Vol. 119, edited by H. D. III and A. Singh (PMLR, 2020) pp. 1597–1607

    work page 2020

  26. [26]

    I. Cong, S. Choi, and M. D. Lukin, Quantum convolu- tional neural networks, Nature Physics 15, 1273 (2019)

    work page 2019

  27. [27]

    Chinzei, Q

    K. Chinzei, Q. H. Tran, K. Maruyama, H. Oshima, and S. Sato, Splitting and parallelizing of quantum convolu- tional neural networks for learning translationally sym- metric data, Phys. Rev. Res. 6, 023042 (2024)

    work page 2024

  28. [28]

    Resource-efficient equivariant quantum convolutional neural networks

    K. Chinzei, Q. H. Tran, Y. Endo, and H. Oshima, Resource-efficient equivariant quantum convolutional neural networks, arXiv:2410.01252 [quant-ph] (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  29. [29]

    P´ erez-Salinas, A

    A. P´ erez-Salinas, A. Cervera-Lierta, E. Gil-Fuster, and J. I. Latorre, Data re-uploading for a universal quantum classifier, Quantum 4, 226 (2020), 1907.02085

    work page arXiv 2020

  30. [30]

    X. Gao, E. R. Anschuetz, S.-T. Wang, J. I. Cirac, and M. D. Lukin, Enhancing generative models via quantum correlations, Phys. Rev. X 12, 021037 (2022)

    work page 2022

  31. [31]

    M. S. Rudolph, N. B. Toussaint, A. Katabarwa, S. Johri, B. Peropadre, and A. Perdomo-Ortiz, Generation of high- resolution handwritten digits with an ion-trap quantum computer, Phys. Rev. X 12, 031010 (2022)

    work page 2022

  32. [32]

    Van Den Nest, Simulating quantum computers with probabilistic methods, Quantum Info

    M. Van Den Nest, Simulating quantum computers with probabilistic methods, Quantum Info. Comput. 11, 784–812 (2011)

    work page 2011

  33. [33]

    Schuld, V

    M. Schuld, V. Bergholm, C. Gogolin, J. Izaac, and N. Kil- loran, Evaluating analytic gradients on quantum hard- ware, Phys. Rev. A 99, 032331 (2019)

    work page 2019

  34. [34]

    Huang, Q.-Y

    G.-B. Huang, Q.-Y. Zhu, and C.-K. Siew, Extreme learn- ing machine: Theory and applications, Neurocomputing 70, 489 (2006)

    work page 2006

  35. [35]

    Mujal, R

    P. Mujal, R. Mart´ ınez-Pe˜ na, J. Nokkala, J. Garc´ ıa-Beni, G. L. Giorgi, M. C. Soriano, and R. Zambrini, Opportu- nities in quantum reservoir computing and extreme learn- ing machines, Adv. Quantum Technol. , 2100027 (2021)

    work page 2021

  36. [36]

    Fujii and K

    K. Fujii and K. Nakajima, Harnessing disordered- ensemble quantum dynamics for machine learning, Phys. 13 Rev. Applied 8, 024030 (2017)

    work page 2017

  37. [37]

    Nakajima, K

    K. Nakajima, K. Fujii, M. Negoro, K. Mitarai, and M. Kitagawa, Boosting computational power through spatial multiplexing in quantum reservoir computing, Phys. Rev. Applied 11, 034021 (2019)

    work page 2019

  38. [38]

    Q. H. Tran and K. Nakajima, Learning temporal quan- tum tomography, Phys. Rev. Lett. 127, 260401 (2021)

    work page 2021

  39. [39]

    Kubota, Y

    T. Kubota, Y. Suzuki, S. Kobayashi, Q. H. Tran, N. Ya- mamoto, and K. Nakajima, Temporal information pro- cessing induced by quantum noise, Phys. Rev. Res. 5, 023057 (2023)

    work page 2023

  40. [40]

    Q. H. Tran and K. Nakajima, Higher-order quan- tum reservoir computing, Preprint at arXiv:2006.08999 (2020)

    work page arXiv 2006

  41. [41]

    Momeni, B

    A. Momeni, B. Rahmani, M. Mall´ ejac, P. del Hougne, and R. Fleury, Backpropagation-free training of deep physical neural networks, Science382, 1297–1303 (2023)

    work page 2023

  42. [42]

    D. Wu, R. Rossi, F. Vicentini, N. Astrakhantsev, F. Becca, X. Cao, J. Carrasquilla, F. Ferrari, A. Georges, M. Hibat-Allah, M. Imada, A. M. L¨ auchli, G. Mazzola, A. Mezzacapo, A. Millis, J. Robledo Moreno, T. Neupert, Y. Nomura, J. Nys, O. Parcollet, R. Pohle, I. Romero, M. Schmid, J. M. Silvester, S. Sorella, L. F. Tocchio, L. Wang, S. R. White, A. Wiet...

    work page 2024

  43. [43]

    Verresen, R

    R. Verresen, R. Moessner, and F. Pollmann, One- dimensional symmetry protected topological phases and their transitions, Phys. Rev. B 96, 165124 (2017)

    work page 2017

  44. [44]

    Schuch, D

    N. Schuch, D. P´ erez-Garc´ ıa, and I. Cirac, Classifying quantum phases using matrix product states and pro- jected entangled pair states, Phys. Rev. B 84, 165139 (2011)

    work page 2011

  45. [45]

    Recio-Armengol, F

    E. Recio-Armengol, F. J. Schreiber, J. Eisert, and C. Bravo-Prieto, Learning complexity gradu- ally in quantum machine learning models, arXiv 10.48550/arXiv.2411.11954 (2024)

    work page doi:10.48550/arxiv.2411.11954 2024

  46. [46]

    H. Wang, Y. Ding, J. Gu, Y. Lin, D. Z. Pan, F. T. Chong, and S. Han, Quantumnas: Noise- adaptive search for robust quantum circuits, in Proc. 28th IEEE Int. Symp. High-Perform. Comput. Arch., HCA ’28 (2022)

    work page 2022

  47. [47]

    UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction

    L. McInnes, J. Healy, and J. Melville, Umap: Uniform manifold approximation and projection for dimension re- duction, arXiv 10.48550/arXiv.1802.03426 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1802.03426 2018

  48. [48]

    H. Xiao, K. Rasul, and R. Vollgraf, Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms, arXiv 10.48550/arXiv.1708.07747 (2017)

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1708.07747 2017

  49. [49]

    K. Zha, P. Cao, J. Son, Y. Yang, and D. Katabi, Rank-n-contrast: learning con- tinuous representations for regression, in Proc. 37th Int. Conf. on Neural Info. Process. Systems, NIPS ’23 (Curran Associates Inc., Red Hook, NY, USA, 2023)

    work page 2023

  50. [50]

    Nøkland, Direct feedback alignment pro- vides learning in deep neural networks, in Proc

    A. Nøkland, Direct feedback alignment pro- vides learning in deep neural networks, in Proc. 30th Int. Conf. on Neural Info. Process. Systems, NIPS’16 (Curran Associates Inc., Red Hook, NY, USA,

    work page

  51. [51]

    Nakajima, K

    M. Nakajima, K. Inoue, K. Tanaka, Y. Kuniyoshi, T. Hashimoto, and K. Nakajima, Physical deep learning with biologically inspired training method: gradient-free approach for physical hardware, Nat. Commun. 13, 7847 (2022)

    work page 2022

  52. [52]

    Q. Li, Y. W. Teh, and R. Pascanu, Noprop: Training neural networks without back-propagation or forward- propagation, arXiv 10.48550/arXiv.2503.24322 (2025)

    work page doi:10.48550/arxiv.2503.24322 2025

  53. [53]

    D. P. Kingma and J. Ba, Adam: A method for stochastic optimization, in Proc. 3rd Int. Conf. on Learning Representations, ICLR ’15 (2015)

    work page 2015