pith. sign in

arxiv: 2602.18377 · v3 · submitted 2026-02-20 · 🪐 quant-ph · cs.LG

Theory and interpretability of Quantum Extreme Learning Machines: a Pauli-transfer matrix approach

Markus Gross , Hans-Martin Rieser This is my paper

Pith reviewed 2026-05-15 20:23 UTC · model grok-4.3

classification 🪐 quant-ph cs.LG
keywords quantum extreme learning machinesPauli transfer matrixquantum reservoir computingnonlinear dynamical systemsflow map approximationoperator spreadingreadout rankinterpretable quantum models
0
0 comments X

The pith

The Pauli transfer matrix formalism converts quantum extreme learning machines into interpretable nonlinear vector regression models that approximate dynamical flow maps.

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

This paper applies the Pauli transfer matrix formalism to quantum extreme learning machines that use initial-state encoding and continuous-time unitary reservoir dynamics. The formalism shows that the encoding step generates a full set of nonlinear Pauli features, which the subsequent quantum channels transform linearly before the chosen measurements extract them for the readout. This structure lets the entire QELM be rewritten as a classical nonlinear vector autoregression model. When the machine is trained on trajectories of nonlinear dynamical systems, the resulting model acts as a surrogate approximation to the system's underlying flow map. A reader would care because the approach replaces black-box quantum training with an explicit classical description that reveals how encoding, dynamics, and measurements together determine what the machine can learn.

Core claim

The PTM formalism yields a nonlinear vector (auto-)regression model as an interpretable classical representation of a QELM. Operator spreading under unitary evolution determines decodability of the Pauli features generated by encoding. Structured Hamiltonians reduce readout rank when their symmetries align with the chosen observables. In the forecasting application, a QELM trained on trajectories learns a surrogate approximation to the underlying flow map of the dynamical system.

What carries the argument

The Pauli transfer matrix representation of the quantum channel, which linearly maps the nonlinear Pauli features produced by the encoding step.

If this is right

  • Optimizing a QELM reduces to choosing measurement operators that decode task-relevant features after the channel transformations have been applied.
  • The spreading of operators under the unitary evolution sets which nonlinear features remain accessible to the regressor.
  • Hamiltonian symmetries produce low-rank readouts for certain observable families, directly limiting model expressivity.
  • Temporal multiplexing of measurements can be analyzed as additional linear combinations that enlarge the set of decodable features.

Where Pith is reading between the lines

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

  • If the PTM reduction holds, one could deliberately select Hamiltonians lacking the identified symmetries to raise readout rank and increase capacity for harder forecasting tasks.
  • The classical nonlinear regression form opens the possibility of importing standard regularization or feature-selection methods from statistics to improve QELM training.
  • The same framework might be used to predict how modest amounts of decoherence would degrade the learned flow-map approximation, providing a concrete target for hardware experiments.

Load-bearing premise

The continuous-time unitary reservoir dynamics and chosen measurement operators allow the PTM to fully capture all relevant nonlinear features without decoherence, noise, or finite-sampling effects altering the effective transformations.

What would settle it

A direct comparison, on the same training trajectories, between the PTM-predicted feature transformations and the actual measurement statistics obtained from the quantum device; any systematic mismatch would show that the classical regression model fails to represent the QELM.

Figures

Figures reproduced from arXiv: 2602.18377 by Hans-Martin Rieser, Markus Gross.

Figure 1
Figure 1. Figure 1: Time evolution of the Pauli weight average ¯ν [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Visualization of the operator spreading (4.12) via the effective PTM [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Projector R +R of the effective PTM [Eq. (4.1)] for different unitary types and a complete weight-1 Pauli observable set. Panels (a,b) directly correspond to the R shown in [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Feature decodability score ¯γ 2 j obtained from the effective PTM R [Eq. (4.18)], averaged over each Pauli weight sector, of a n = 3 qubit model as a function of temporal multiplexing iteration length L [Eq. (4.1)]. The unitary evolution time is given by tL = L (i.e., initial time t1 = 1). The corresponding rank and number of rows of R, representing the measurement budget (B = mL), are shown on the top axi… view at source ↗
Figure 5
Figure 5. Figure 5: Rank of the effective PTM R [Eq. (4.1)] obtain by temporal multiplexing an observable set S (one and two-site Pauli Z) under unitary evolution with a TFIM Hamiltonian [Eq. (2.5)]. (a) Evolution of rank(R) with temporal multiplexing steps L for the TFIM (2.5a) and S = {Zi, ZiZj} observables. (b) Asymptotic PTM rank, limL→∞ rank(R), obtained numerically (solid curves), compared to the theoretical scaling pre… view at source ↗
Figure 6
Figure 6. Figure 6: Ideal nonlinear capacity score R2 (k) [Eq. (4.21)] for amplitude and rotational encodings [Eqs. (2.8) and (2.10)], indicating the ability of an ideal QELM f(u) = w⊤ϕ(u) to generate degree-k monomials of the input uα (α = 1, . . . , D and n = D). Unless indicated, only Pauli-Z features are measured in (a). Nonlinear capacity is significantly enhanced when measuring over all Paulis (b). This is also demonstr… view at source ↗
Figure 7
Figure 7. Figure 7: Nonlinear capacity score R2 (k) [Eq. (4.21)] as a function of the nonlinear order k and the temporal multiplexing iteration length L [Eq. (4.1)]. The unitary evolution time is given by tL = L. We consider a n = 3 qubit model in the same setup as in [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Relative nonlinear capacity score R2 (k)/(R2 all(k) + δ) [Eq. (4.21)] as a function of the nonlinear order k and the temporal multiplexing iteration length L [Eq. (4.1)]. R2 all is the nonlinear capacity score obtained for a complete Pauli observable set, and we choose δ = 0.1 as a cut-off since scores R2 all ≲ 0.1 are not considered meaningful. We use the same setup as in [PITH_FULL_IMAGE:figures/full_fi… view at source ↗
Figure 9
Figure 9. Figure 9: Ability to construct monomials of the input [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: QELM vs. its classical representation: The forecast obtained from a QELM [Eq. (2.1)] trained [PITH_FULL_IMAGE:figures/full_fig_p021_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Prediction performance (forecast horizon and RMS training error) under temporal multiplexing [PITH_FULL_IMAGE:figures/full_fig_p022_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Illustration of the flow map learned by a QELM, obtained from the Taylor-approximation [PITH_FULL_IMAGE:figures/full_fig_p023_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Illustration of the flow map learned by a QELM, obtained from the Taylor-approximation [PITH_FULL_IMAGE:figures/full_fig_p024_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Pauli weight average (a) of Heisenberg-evolved measurement observables can predict the training [PITH_FULL_IMAGE:figures/full_fig_p026_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Null-space criterion (B1): Number of isolatable features as a function of the number of observ [PITH_FULL_IMAGE:figures/full_fig_p028_15.png] view at source ↗
read the original abstract

Quantum reservoir computers (QRCs) have emerged as a promising approach to quantum machine learning, since they utilize the natural dynamics of quantum systems for data processing and are simple to train. Here, we consider $n$-qubit quantum extreme learning machines (QELMs) with initial-state encoding and continuous-time reservoir dynamics. We apply the Pauli transfer matrix (PTM) formalism to theoretically analyze the influence of encoding, reservoir dynamics, and measurement operations (including temporal multiplexing) on the QELM performance. This formalism reveals the complete set of (nonlinear) features generated by the encoding, and shows how the subsequent quantum channels linearly transform these Pauli features before they are probed by the chosen measurement operators. Optimizing such a QELM can therefore be cast as a decoding problem in which one shapes the channel-induced transformations such that task-relevant features become available to the regressor, effectively reversing the information scrambling of a unitary. Operator spreading under unitary evolution determines decodability of Pauli features, which underlies the nonlinear processing capacity of the reservoir. When paired with certain observables, structured Hamiltonians can reduce model expressivity, as reflected in a low readout rank. We trace this effect to Hamiltonian symmetries and derive asymptotic rank estimates for symmetry-resolved observable families. The PTM formalism yields a nonlinear vector (auto-)regression model as an interpretable classical representation of a QELM. As a specific application, we focus on forecasting nonlinear dynamical systems and show that a QELM trained on such trajectories learns a surrogate-approximation to the underlying flow map.

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

2 major / 2 minor

Summary. The manuscript develops a Pauli transfer matrix (PTM) formalism for analyzing n-qubit quantum extreme learning machines (QELMs) with initial-state encoding and continuous-time unitary reservoir dynamics. It shows that encoding generates a complete set of nonlinear Pauli features that are linearly transformed by the quantum channels before being probed by measurement operators (including temporal multiplexing). This yields an explicit nonlinear vector autoregression model as the classical interpretable representation of the QELM. The formalism is applied to forecasting nonlinear dynamical systems, where the trained QELM learns a surrogate approximation to the underlying flow map. Symmetry-resolved rank bounds for observables are derived from Hamiltonian symmetries.

Significance. If the central derivations hold, the work provides a rigorous algebraic bridge between quantum reservoir dynamics and classical nonlinear regression, enabling interpretability and design guidance for QELMs. The exact equivalence in the ideal unitary, infinite-shot limit (with no hidden parameters or fitted quantities) and the symmetry-based rank estimates are notable strengths that could inform practical reservoir engineering for time-series tasks.

major comments (2)
  1. [§4] §4 (PTM-derived regression model): the mapping to the nonlinear vector autoregression is derived algebraically for the ideal case, but the manuscript should explicitly bound the deviation when finite-shot sampling or weak decoherence is present, as these alter the effective channel transformations and could affect the surrogate flow-map claim for dynamical systems.
  2. [§5.2] §5.2 (dynamical systems forecasting): the surrogate-approximation result is shown via training on trajectories, but the rank estimates for symmetry-resolved observables (derived in §3) are not quantitatively linked to the observed prediction accuracy; a direct comparison of readout rank versus forecast error would strengthen the expressivity analysis.
minor comments (2)
  1. [Throughout] Notation for the PTM elements and feature vectors should be unified across sections to avoid redefinition of symbols such as the encoding map and channel superoperators.
  2. [Figure captions] Figure captions for the dynamical systems examples should include the specific Hamiltonian and observable choices used to generate the rank bounds.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive report and positive evaluation of the work. We address each major comment below and describe the revisions we will implement in the next version of the manuscript.

read point-by-point responses

  1. Referee: [§4] §4 (PTM-derived regression model): the mapping to the nonlinear vector autoregression is derived algebraically for the ideal case, but the manuscript should explicitly bound the deviation when finite-shot sampling or weak decoherence is present, as these alter the effective channel transformations and could affect the surrogate flow-map claim for dynamical systems.

    Authors: We agree that explicit bounds on deviations from the ideal unitary, infinite-shot limit would strengthen the practical interpretation. In the revised manuscript we will add a short subsection to §4 that derives first-order perturbative bounds: shot-noise deviations via Hoeffding-type concentration on the empirical PTM entries, and weak decoherence via a first-order expansion in the Lindblad operators. These bounds will be connected to the resulting error in the learned surrogate flow map for the dynamical-systems examples. revision: yes

  2. Referee: [§5.2] §5.2 (dynamical systems forecasting): the surrogate-approximation result is shown via training on trajectories, but the rank estimates for symmetry-resolved observables (derived in §3) are not quantitatively linked to the observed prediction accuracy; a direct comparison of readout rank versus forecast error would strengthen the expressivity analysis.

    Authors: We accept this suggestion. The original text presents the symmetry-resolved rank bounds in §3 but does not numerically correlate them with forecasting performance. In the revision we will augment §5.2 with a new figure that reports the effective readout rank (computed for each symmetry class of observables) alongside the corresponding mean forecast error on the tested dynamical systems, thereby providing a direct quantitative link between the theoretical rank analysis and the observed accuracy. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected in PTM derivation

full rationale

The paper applies the standard Pauli transfer matrix formalism as an external analytical tool to map QELM encoding, continuous-time unitary dynamics, and measurements onto an explicit nonlinear vector autoregression model. All load-bearing steps are algebraic identities that hold exactly in the stated ideal unitary, infinite-shot limit; the surrogate flow-map claim follows directly from training the resulting classical regressor on trajectory data without any reduction to fitted parameters, self-definitional loops, or load-bearing self-citations. No uniqueness theorems or ansatzes are imported from the authors' prior work, and symmetry-based rank bounds are derived directly from the PTM representation itself. The derivation is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The analysis rests on standard quantum mechanics assumptions for unitary evolution and the applicability of the PTM formalism; no new free parameters or invented entities are introduced in the abstract.

axioms (2)
  • domain assumption The n-qubit reservoir evolves under continuous-time unitary dynamics generated by a time-independent Hamiltonian.
    Invoked to define the quantum channel acting on the encoded Pauli features.
  • domain assumption Measurements are performed via expectation values of chosen observables, possibly with temporal multiplexing.
    Required to probe the transformed Pauli features after evolution.

pith-pipeline@v0.9.0 · 5579 in / 1376 out tokens · 51628 ms · 2026-05-15T20:23:24.529683+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.

Reference graph

Works this paper leans on

138 extracted references · 138 canonical work pages · 3 internal anchors

  1. [1]

    edge of chaos

    Nonlinear monomials at readout Raw monomials inu α are often relevant features for practical applications, such as autoregressive learning of dynamical systems. Let thusB(u) ={1, u 1, . . . , u1u2, . . . ,Q α ur α}be the vector of size bof monomials of{u α}1≤α≤D up to degreer. The order-rTaylor expansion of the Pauli features {ϕj(u)}1≤j≤q within a small r...

    work page 2000

  2. [2]

    For Clifford unitaries,T U is a signed permutation of Pauli strings

    Unitary channels For a unitary channelU(ρ) =U ρU †, one has TU,αβ = 1 2n tr Pα U PβU † ,(D1) Trace preservation and unitality imply the form TU = 10 ⊤ 0Ω U ! ,(D2) i.e., the traceless coefficients undergo a real orthogonal rotation with Ω U ∈SO(d 2 −1). For Clifford unitaries,T U is a signed permutation of Pauli strings. Consider aseparablereservoir Hamil...

    work page

  3. [3]

    Then TE,αβ = 1 2n X ℓ tr PαKℓPβK † ℓ ,(D7) which corresponds in block form (compatible withϕ= (1,s)) to TE = 10 ⊤ tΩ ! ,s ′ =t+ Ωs,(D8) wheret=0for unital channels

    General quantum channels: affine action on the Bloch part Consider a channelEthat is completely positive and trace-preserving (CPTP) and has associated Kraus operators{K ℓ}. Then TE,αβ = 1 2n X ℓ tr PαKℓPβK † ℓ ,(D7) which corresponds in block form (compatible withϕ= (1,s)) to TE = 10 ⊤ tΩ ! ,s ′ =t+ Ωs,(D8) wheret=0for unital channels. For a single qubit...

    work page

  4. [4]

    (2.5b), we introduce Jordan–Wigner Majorana quasiparticle op- erators γ2j−1 = Y m<j Zm Xj, γ 2j = Y m<j Zm Yj, j= 1,

    Heisenberg sectors of the relevant observables For the TFIM in the form Eq. (2.5b), we introduce Jordan–Wigner Majorana quasiparticle op- erators γ2j−1 = Y m<j Zm Xj, γ 2j = Y m<j Zm Yj, j= 1, . . . , n,(E1) resulting in Hxx-z TFIM =iJ n−1X j=1 γ2jγ2j+1 +ih nX j=1 γ2j−1γ2j.(E2) Since the Hamiltonian is quadratic in the Majoranas, Heisenberg evolution acts...

    work page

  5. [5]

    Denote byR L the effective PTM in Eq

    Asymptotic rank of the temporally multiplexed PTM We now connect the above Heisenberg-sector structure to the effective PTM in temporal mul- tiplexing. Denote byR L the effective PTM in Eq. (4.1) evaluated forLmultiplexing steps and 34 L(·) =i[H,·]. For generic couplingsJ, h̸= 0 and generic sampling times{t ℓ}, the large-Lrank is the dimension of the Heis...

    work page

  6. [6]

    A HamiltonianHinduces the Heisenberg evolutionO (H)(t) =e tLOwith LiouvillianL(O) =i[H, O]

    Operator-space formulation of the effective PTM LetHbe ad-dimensional Hilbert space and letObe the associated operator space equipped with the Hilbert–Schmidt inner product⟨A, B⟩= 1 dtr(A†B). A HamiltonianHinduces the Heisenberg evolutionO (H)(t) =e tLOwith LiouvillianL(O) =i[H, O]. Given a selected observable familyS, the “seed” space of measurement oper...

    work page

  7. [7]

    Invariant sectors induced by symmetries Assume a symmetry of the Liouvillian described by the superoperator Γ :O → Osatisfying [Γ,L] = 0.(F6) In the physically relevant cases of a unitary or Hermitian Γ, the spectral theorem yields Γ =P α λαΠα, where the Π α are mutually orthogonal spectral projectors. These render the orthogonal decomposition O= M α Oα,Π...

    work page

  8. [8]

    (F16) determines only the maximal accessible dimension

    Hamiltonian criteria for underexploration The symmetry decomposition in Eq. (F16) determines only the maximal accessible dimension. Whether a populated sector is fully explored depends on the restricted generatorL| Oα and on the choice of seed subspaceM α. LetO α be a symmetry-compatible invariant sector andL α :=L| Oα. As above, we consider the spectral ...

    work page

  9. [9]

    We set thusF=Oand choose a one-dimensional seed spaceM= span{O}for a single observableO

    Single-observable limit We now apply the above formalism, to the Krylov grade of a single observable, recovering a key result in [47]. We set thusF=Oand choose a one-dimensional seed spaceM= span{O}for a single observableO. Then the Krylov gradeMis the dimension of the minimal Liouvillian Krylov space generated byO[47], M:= dim span{O,L(O),L 2(O), . . .}=...

    work page

  10. [10]

    Hilbert-space symmetries Consider an ordinary Hilbert-space symmetries withU HU † =H

    Examples of symmetry-induced rank bounds a. Hilbert-space symmetries Consider an ordinary Hilbert-space symmetries withU HU † =H. Then the superoperator ΓU(O) =U OU † (F29) is unitary onOand satisfies [Γ U ,L] = 0. Similarly, for aZ 2 symmetry withU 2 =I, the corre- sponding superoperator is Π± = 1 2(I O ±Γ U).(F30) IfQ= P q qPq is a conserved charge with...

    work page

  11. [11]

    Consequences for PTM analysis and feature decodability In a symmetry-adapted orthonormal operator basis, the unitary PTM becomes block diagonal, V(t) = M α Vα(t),(F36) Accordingly, the temporally multiplexed effective PTM (4.1) can be brought into a block form RL ∼ M α RL,α,(F37) and the projector defined in Eq. (4.18) decomposes as R+ L RL ∼ M α R+ L,αRL...

    work page

  12. [12]

    Fujii and K

    K. Fujii and K. Nakajima, Harnessing Disordered-Ensemble Quantum Dynamics for Machine Learn- ing, Phys. Rev. Applied8, 024030 (2017). 40

    work page 2017

  13. [13]

    Sakurai, M

    A. Sakurai, M. P. Estarellas, W. J. Munro, and K. Nemoto, Quantum Extreme Reservoir Computation Utilizing Scale-Free Networks, Phys. Rev. Applied17, 064044 (2022)

    work page 2022

  14. [14]

    Kornjača, H.-Y

    M. Kornjaˇ ca, H.-Y. Hu, C. Zhao, J. Wurtz, P. Weinberg, M. Hamdan, A. Zhdanov, S. H. Cantu, H. Zhou, R. A. Bravo, K. Bagnall, J. I. Basham, J. Campo, A. Choukri, R. DeAngelo, P. Frederick, D. Haines, J. Hammett, N. Hsu, M.-G. Hu, F. Huber, P. N. Jepsen, N. Jia, T. Karolyshyn, M. Kwon, J. Long, J. Lopatin, A. Lukin, T. Macr` ı, O. Markovi´ c, L. A. Mart´ ...

    work page arXiv 2024

  15. [15]

    Sakurai, A

    A. Sakurai, A. Hayashi, W. J. Munro, and K. Nemoto, Simple Hamiltonian dynamics is a powerful quantum processing resource (2025), arXiv:2405.14245 [quant-ph]

    work page arXiv 2025

  16. [16]

    A. D. Lorenzis, M. P. Casado, M. P. Estarellas, N. L. Gullo, T. Lux, F. Plastina, A. Riera, and J. Settino, Harnessing Quantum Extreme Learning Machines for image classification, Phys. Rev. Applied23, 044024 (2025), arXiv:2409.00998 [quant-ph]

    work page arXiv 2025

  17. [17]

    J. Chen, H. I. Nurdin, and N. Yamamoto, Temporal Information Processing on Noisy Quantum Computers, Phys. Rev. Applied14, 024065 (2020)

    work page 2020

  18. [18]

    Suzuki, Q

    Y. Suzuki, Q. Gao, K. C. Pradel, K. Yasuoka, and N. Yamamoto, Natural quantum reservoir com- puting for temporal information processing, Scientific Reports12, 1353 (2022)

    work page 2022

  19. [19]

    Mujal, R

    P. Mujal, R. Mart´ ınez-Pe˜ na, G. L. Giorgi, M. C. Soriano, and R. Zambrini, Time Series Quantum Reservoir Computing with Weak and Projective Measurements, npj Quantum Information9, 16 (2023), arXiv:2205.06809 [quant-ph]

    work page arXiv 2023

  20. [20]

    Ahmed, F

    O. Ahmed, F. Tennie, and L. Magri, Prediction of chaotic dynamics and extreme events: A recurrence- free quantum reservoir computing approach (2024), arXiv:2405.03390 [quant-ph]

    work page arXiv 2024

  21. [21]

    Ahmed, F

    O. Ahmed, F. Tennie, and L. Magri, Optimal training of finitely-sampled quantum reservoir computers for forecasting of chaotic dynamics (2024), arXiv:2409.01394 [quant-ph]

    work page arXiv 2024

  22. [22]

    Pfeffer, F

    P. Pfeffer, F. Heyder, and J. Schumacher, Hybrid quantum-classical reservoir computing of thermal convection flow, Phys. Rev. Research4, 033176 (2022), arXiv:2204.13951 [quant-ph]

    work page arXiv 2022

  23. [23]

    Pfeffer, F

    P. Pfeffer, F. Heyder, and J. Schumacher, Reduced-order modeling of two-dimensional turbulent Rayleigh-B´ enard flow by hybrid quantum-classical reservoir computing, Phys. Rev. Research5, 043242 (2023), arXiv:2307.03053 [physics]

    work page arXiv 2023

  24. [24]

    Suprano, D

    A. Suprano, D. Zia, L. Innocenti, S. Lorenzo, V. Cimini, T. Giordani, I. Palmisano, E. Polino, N. Spagnolo, F. Sciarrino, G. M. Palma, A. Ferraro, and M. Paternostro, Experimental Property Reconstruction in a Photonic Quantum Extreme Learning Machine, Phys. Rev. Lett.132, 160802 (2024)

    work page 2024

  25. [25]

    Vetrano, G

    M. Vetrano, G. Lo Monaco, L. Innocenti, S. Lorenzo, and G. M. Palma, State estimation with quantum extreme learning machines beyond the scrambling time, npj Quantum Information11, 20 (2025)

    work page 2025

  26. [26]

    D. Zia, L. Innocenti, G. Minati, S. Lorenzo, A. Suprano, R. D. Bartolo, N. Spagnolo, T. Giordani, V. Cimini, G. M. Palma, A. Ferraro, F. Sciarrino, and M. Paternostro, Quantum reservoir com- puting for photonic entanglement witnessing, Science Advances11, 10.1126/sciadv.ady7987 (2025), arXiv:2502.18361 [quant-ph]

    work page doi:10.1126/sciadv.ady7987 2025

  27. [27]

    Assil, A

    H. Assil, A. El Allati, and G. L. Giorgi, Entanglement estimation of Werner states with a quantum extreme learning machine, Phys. Rev. A111, 022412 (2025)

    work page 2025

  28. [28]

    J. R. McClean, S. Boixo, V. N. Smelyanskiy, R. Babbush, and H. Neven, Barren plateaus in quantum neural network training landscapes, Nature Communications9, 4812 (2018)

    work page 2018

  29. [29]

    Larocca, S

    M. Larocca, S. Thanasilp, S. Wang, K. Sharma, J. Biamonte, P. J. Coles, L. Cincio, J. R. McClean, Z. Holmes, and M. Cerezo, Barren Plateaus in Variational Quantum Computing, Nature Reviews Physics7, 174 (2025), arXiv:2405.00781 [quant-ph]

    work page arXiv 2025

  30. [30]

    Domingo, G

    L. Domingo, G. Carlo, and F. Borondo, Taking advantage of noise in quantum reservoir computing (2023), arXiv:2301.06814 [quant-ph]

    work page arXiv 2023

  31. [31]

    D. Fry, A. Deshmukh, S. Y.-C. Chen, V. Rastunkov, and V. Markov, Optimizing quantum noise- induced reservoir computing for nonlinear and chaotic time series prediction, Scientific Reports13, 19326 (2023). 41

    work page 2023

  32. [32]

    Sannia, R

    A. Sannia, R. Mart´ ınez-Pe˜ na, M. C. Soriano, G. L. Giorgi, and R. Zambrini, Dissipation as a resource for Quantum Reservoir Computing, Quantum8, 1291 (2024)

    work page 2024

  33. [33]

    Monzani, E

    F. Monzani, E. Ricci, L. Nigro, and E. Prati, Non-unital noise in a superconducting quantum computer as a computational resource for reservoir computing (2024), arXiv:2409.07886 [quant-ph]

    work page arXiv 2024

  34. [34]

    Mujal, J

    P. Mujal, J. Nokkala, R. Mart´ ınez-Pe˜ na, G. L. Giorgi, M. C. Soriano, and R. Zambrini, Analytical evidence of nonlinearity in qubits and continuous-variable quantum reservoir computing, Journal of Physics: Complexity2, 045008 (2021)

    work page 2021

  35. [35]

    Nokkala, R

    J. Nokkala, R. Mart´ ınez-Pe˜ na, G. L. Giorgi, V. Parigi, M. C. Soriano, and R. Zambrini, Gaussian states of continuous-variable quantum systems provide universal and versatile reservoir computing, Communications Physics4, 53 (2021)

    work page 2021

  36. [36]

    L. C. G. Govia, G. J. Ribeill, G. E. Rowlands, and T. A. Ohki, Nonlinear input transformations are ubiquitous in quantum reservoir computing, Neuromorphic Computing and Engineering2, 014008 (2022)

    work page 2022

  37. [37]

    Chen and H

    J. Chen and H. I. Nurdin, Learning Nonlinear Input-Output Maps with Dissipative Quantum Systems, Quantum Information Processing18, 198 (2019), arXiv:1901.01653 [quant-ph]

    work page arXiv 2019

  38. [38]

    T. Goto, Q. H. Tran, and K. Nakajima, Universal Approximation Property of Quantum Machine Learning Models in Quantum-Enhanced Feature Spaces, Phys. Rev. Lett.127, 090506 (2021), arXiv:2009.00298 [quant-ph]

    work page arXiv 2021

  39. [39]

    Mart´ ınez-Pe˜ na and J.-P

    R. Mart´ ınez-Pe˜ na and J.-P. Ortega, Quantum reservoir computing in finite dimensions, Phys. Rev. E 107, 035306 (2023)

    work page 2023

  40. [40]

    Monzani and E

    F. Monzani and E. Prati, Universality conditions of unified classical and quantum reservoir computing (2024), arXiv:2401.15067 [quant-ph]

    work page arXiv 2024

  41. [41]

    Gonon and A

    L. Gonon and A. Jacquier, Universal Approximation Theorem and Error Bounds for Quantum Neural Networks and Quantum Reservoirs, IEEE Transactions on Neural Networks and Learning Systems 36, 11355 (2025)

    work page 2025

  42. [42]

    Schuld, Supervised quantum machine learning models are kernel methods, Journal of Machine Learning Research 10.1145/3451902.3451903 (2021)

    M. Schuld, Supervised quantum machine learning models are kernel methods, Journal of Machine Learning Research 10.1145/3451902.3451903 (2021)

    work page doi:10.1145/3451902.3451903 2021

  43. [43]

    Gross and H.-M

    M. Gross and H.-M. Rieser, Kernel-based optimization of measurement operators for quantum reser- voir computers (2026), arXiv:2602.14677 [quant-ph]

    work page arXiv 2026

  44. [44]

    Canatar, E

    A. Canatar, E. Peters, C. Pehlevan, and S. Wild, Bandwidth enables generalization in quantum kernel models, Trans. Mach. Learn. Res.2023(2022)

    work page 2023

  45. [45]

    Gross, M

    M. Gross, M. Lange, B. Bujnowski, and H.-M. Rieser, Expressivity vs. Generalization in Quantum Kernel Methods, in33rd European Symposium on Artificial Neural Networks, Computational Intel- ligence and Machine Learning, ESANN 2025, edited by M. Verleysen (Ciaco - i6doc.com, Bruges, Belgium, 2025) pp. 543–548

    work page 2025

  46. [46]

    Kempkes, A

    M. Kempkes, A. Ijaz, E. Gil-Fuster, C. Bravo-Prieto, J. Spiegelberg, E. van Nieuwenburg, and V. Dun- jko, Double descent in quantum kernel methods, PRX Quantum7, 010312 (2026), arXiv:2501.10077 [quant-ph]

    work page arXiv 2026

  47. [47]

    Fujii and K

    K. Fujii and K. Nakajima, Quantum reservoir computing: A reservoir approach toward quantum machine learning on near-term quantum devices (2020), arXiv:2011.04890 [quant-ph]

    work page arXiv 2020

  48. [48]

    Mujal, R

    P. Mujal, R. Mart´ ınez-Pe˜ na, J. Nokkala, J. Garc´ ıa-Beni, G. L. Giorgi, M. C. Soriano, and R. Zam- brini, Opportunities in Quantum Reservoir Computing and Extreme Learning Machines, Advanced Quantum Technologies4, 2100027 (2021)

    work page 2021

  49. [49]

    C. M. Wilson, J. S. Otterbach, N. Tezak, R. S. Smith, A. M. Polloreno, P. J. Karalekas, S. Heidel, M. S. Alam, G. E. Crooks, and M. P. da Silva, Quantum Kitchen Sinks: An algorithm for machine learning on near-term quantum computers (2019), arXiv:1806.08321 [quant-ph]

    work page arXiv 2019

  50. [50]

    Huang, Q.-Y

    G.-B. Huang, Q.-Y. Zhu, and C.-K. Siew, Extreme learning machine: Theory and applications, Neu- rocomputing Neural Networks,70, 489 (2006)

    work page 2006

  51. [51]

    Rahimi and B

    A. Rahimi and B. Recht, Random Features for Large-Scale Kernel Machines, inAdvances in Neural Information Processing Systems, Vol. 20 (Curran Associates, Inc., 2007)

    work page 2007

  52. [52]

    Mart´ ınez-Pe˜ na, G

    R. Mart´ ınez-Pe˜ na, G. L. Giorgi, J. Nokkala, M. C. Soriano, and R. Zambrini, Dynamical Phase Transitions in Quantum Reservoir Computing, Phys. Rev. Lett.127, 100502 (2021). 42

    work page 2021

  53. [53]

    Domingo, G

    L. Domingo, G. Carlo, and F. Borondo, Optimal quantum reservoir computing for the noisy intermediate-scale quantum era, Phys. Rev. E106, L043301 (2022)

    work page 2022

  54. [54]

    Hayashi, A

    A. Hayashi, A. Sakurai, S. Nishio, W. J. Munro, and K. Nemoto, Impact of the form of weighted networks on the quantum extreme reservoir computation, Phys. Rev. A108, 042609 (2023), arXiv:2211.07841 [quant-ph]

    work page arXiv 2023

  55. [55]

    G¨ otting, F

    N. G¨ otting, F. Lohof, and C. Gies, Exploring quantumness in quantum reservoir computing, Phys. Rev. A108, 052427 (2023)

    work page 2023

  56. [56]

    Innocenti, S

    L. Innocenti, S. Lorenzo, I. Palmisano, A. Ferraro, M. Paternostro, and G. M. Palma, Potential and limitations of quantum extreme learning machines, Communications Physics6, 118 (2023)

    work page 2023

  57. [57]

    Kobayashi, Q

    S. Kobayashi, Q. H. Tran, and K. Nakajima, Coherence influx is indispensable for quantum reservoir computing (2024), arXiv:2409.12693 [quant-ph]

    work page arXiv 2024

  58. [58]

    ˇCindrak, L

    S. ˇCindrak, L. Jaurigue, and K. L¨ udge, Engineering Quantum Reservoirs through Krylov Complexity, Expressivity and Observability (2025), arXiv:2409.12079 [quant-ph]

    work page arXiv 2025

  59. [59]

    Expressiv- ity limits of quantum reservoir computing, 2025

    N.-E. Sch¨ utte, N. G¨ otting, H. M¨ untinga, M. List, D. Brunner, and C. Gies, Expressivity Limits of Quantum Reservoir Computing (2025), arXiv:2501.15528 [quant-ph]

    work page arXiv 2025

  60. [60]

    Mart´ ınez-Pe˜ na and J.-P

    R. Mart´ ınez-Pe˜ na and J.-P. Ortega, Input-dependence in quantum reservoir computing, Phys. Rev. E111, 065306 (2025)

    work page 2025

  61. [61]

    A. D. Lorenzis, M. P. Casado, N. L. Gullo, T. Lux, F. Plastina, and A. Riera, Entanglement and Classical Simulability in Quantum Extreme Learning Machines (2025), arXiv:2509.06873 [quant-ph]

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  62. [62]

    Schuld, R

    M. Schuld, R. Sweke, and J. J. Meyer, Effect of data encoding on the expressive power of variational quantum-machine-learning models, Phys. Rev. A103, 032430 (2021)

    work page 2021

  63. [63]

    S. Shin, Y. S. Teo, and H. Jeong, Exponential data encoding for quantum supervised learning (2023), arXiv:2206.12105

    work page arXiv 2023

  64. [64]

    Xiong, G

    W. Xiong, G. Facelli, M. Sahebi, O. Agnel, T. Chotibut, S. Thanasilp, and Z. Holmes, On fundamental aspects of quantum extreme learning machines (2023), arXiv:2312.15124 [quant-ph, stat]

    work page arXiv 2023

  65. [65]

    Dambre, D

    J. Dambre, D. Verstraeten, B. Schrauwen, and S. Massar, Information Processing Capacity of Dy- namical Systems, Scientific Reports2, 514 (2012)

    work page 2012

  66. [66]

    Mart´ ınez-Pe˜ na, J

    R. Mart´ ınez-Pe˜ na, J. Nokkala, G. L. Giorgi, R. Zambrini, and M. C. Soriano, Information Processing Capacity of Spin-Based Quantum Reservoir Computing Systems, Cognitive Computation15, 1440 (2023)

    work page 2023

  67. [67]

    When is a system discoverable from data? Discovery requires chaos.ArXiv preprint arXiv:2511.08860, 2025.doi:10.48550/arXiv.2511.08860

    Z. Shumaylov, P. Zaika, P. Scholl, G. Kutyniok, L. Horesh, and C.-B. Sch¨ onlieb, When is a System Discoverable from Data? Discovery Requires Chaos (2025), arXiv:2511.08860 [math]

    work page arXiv 2025

  68. [68]

    McCaul, J

    G. McCaul, J. S. Totero Gongora, W. Otieno, S. Savelev, A. Zagoskin, and A. G. Balanov, Minimal quantum reservoirs with Hamiltonian encoding, Chaos: An Interdisciplinary Journal of Nonlinear Science35, 093135 (2025)

    work page 2025

  69. [69]

    Gaussian Quantum Information

    C. Weedbrook, S. Pirandola, R. Garcia-Patron, N. J. Cerf, T. C. Ralph, J. H. Shapiro, and S. Lloyd, Gaussian Quantum Information, Reviews of Modern Physics84, 621 (2012), arXiv:1110.3234 [quant- ph]

    work page internal anchor Pith review Pith/arXiv arXiv 2012

  70. [70]

    L. C. G. Govia, G. J. Ribeill, G. E. Rowlands, H. K. Krovi, and T. A. Ohki, Quantum reservoir computing with a single nonlinear oscillator, Phys. Rev. Research3, 013077 (2021), arXiv:2004.14965 [quant-ph]

    work page arXiv 2021

  71. [71]

    Schuld and F

    M. Schuld and F. Petruccione,Machine Learning with Quantum Computers(Springer, Cham, 2021)

    work page 2021

  72. [72]

    M. C. Caro, Learning Quantum Processes and Hamiltonians via the Pauli Transfer Matrix, ACM Transactions on Quantum Computing5, 1 (2024), arXiv:2212.04471 [quant-ph]

    work page arXiv 2024

  73. [73]

    Hantzko, L

    L. Hantzko, L. Binkowski, and S. Gupta, Fast generation of Pauli transfer matrices utilizing tensor product structure, Physica Scripta100, 075125 (2025), arXiv:2411.00526 [quant-ph]

    work page arXiv 2025

  74. [74]

    Angrisani, A

    A. Angrisani, A. Schmidhuber, M. S. Rudolph, M. Cerezo, Z. Holmes, and H.-Y. Huang, Classically estimating observables of noiseless quantum circuits (2025), arXiv:2409.01706 [quant-ph]

    work page arXiv 2025

  75. [75]

    H. So, K. W. Chan, Y. Chan, and K. Ho, Linear prediction approach for efficient frequency estimation of multiple real sinusoids: Algorithms and analyses, IEEE Transactions on Signal Processing53, 2290 (2005)

    work page 2005

  76. [76]

    S. V. Vaseghi,Advanced Digital Signal Processing and Noise Reduction(Wiley, Chichester, U.K, 2008). 43

    work page 2008

  77. [77]

    Peters and M

    E. Peters and M. Schuld, Generalization despite overfitting in quantum machine learning models, Quantum6, 777 (2022)

    work page 2022

  78. [78]

    ˇCindrak, B

    S. ˇCindrak, B. Donvil, K. L¨ udge, and L. Jaurigue, Enhancing the performance of quantum reservoir computing and solving the time-complexity problem by artificial memory restriction, Phys. Rev. Research6, 013051 (2024)

    work page 2024

  79. [79]

    ˇCindrak, K

    S. ˇCindrak, K. L¨ udge, and L. Jaurigue, From Krylov Complexity to Observability: Capturing Phase Space Dimension with Applications in Quantum Reservoir Computing (2025), arXiv:2502.12157 [quant-ph]

    work page arXiv 2025

  80. [80]

    Steinegger and C

    J. Steinegger and C. R¨ ath, Predicting three-dimensional chaotic systems with four qubit quantum systems, Scientific Reports15, 6201 (2025), arXiv:2501.15191 [quant-ph]

    work page arXiv 2025

Showing first 80 references.