Recurrent Quantum Feature Maps for Reservoir Computing

Aaron Z. Goldberg; Christoph Simon; Khabat Heshami; Utkarsh Singh

arxiv: 2604.03469 · v1 · submitted 2026-04-03 · 🪐 quant-ph · cs.LG

Recurrent Quantum Feature Maps for Reservoir Computing

Utkarsh Singh , Aaron Z. Goldberg , Christoph Simon , Khabat Heshami This is my paper

Pith reviewed 2026-05-13 18:27 UTC · model grok-4.3

classification 🪐 quant-ph cs.LG

keywords reservoir computingquantum feature mapsrecurrent quantum circuitsMackey-Glass time seriestime series predictionquantum machine learningnoise robustnessmemory capacity

0 comments

The pith

A recurrent quantum feature map reservoir achieves lower error than classical networks on Mackey-Glass time-series prediction with compact qubit and depth requirements.

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

The paper introduces a reservoir computing approach that reuses a fixed quantum circuit to encode both the current input and a classical feedback signal drawn from prior outputs. This recurrent quantum feature map is tested on the Mackey-Glass chaotic time-series forecasting task. It produces lower mean squared error than echo state networks and multilayer perceptrons while keeping circuit depth and qubit count small. Memory capacity analysis shows the model retains temporal information consistent with its accuracy. The construction remains robust to several noise channels but degrades under two-qubit gate errors, marking a concrete near-term constraint.

Core claim

The central claim is that recurrent quantum feature maps formed by feeding a classical signal derived from previous outputs back into a fixed quantum circuit alongside the current input create an effective high-dimensional reservoir. On the Mackey-Glass prediction task this yields lower mean squared error than standard classical baselines while preserving compact resources, with memory capacity measurements confirming retention of temporal structure and noise studies identifying two-qubit gate errors as the dominant limitation.

What carries the argument

Recurrent quantum feature map: a fixed quantum circuit reused to encode both the current input and a classical feedback signal from earlier outputs.

If this is right

Lower mean squared error on Mackey-Glass forecasting than echo state networks or multilayer perceptrons.
Effective retention of temporal information confirmed by explicit memory capacity measurements.
Robustness to several single-qubit noise channels while remaining limited by two-qubit gate errors.
Compact circuit depth and qubit count that fit near-term hardware constraints.

Where Pith is reading between the lines

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

Hybrid feedback loops of this kind could allow quantum reservoirs to handle longer sequences without scaling qubit number.
The same recurrent encoding might extend to other sequential tasks such as speech or sensor streams once two-qubit error rates improve.
Targeted error mitigation focused on two-qubit gates would directly address the identified performance bottleneck.

Load-bearing premise

The classical feedback signal derived from previous outputs can be fed back into the fixed quantum circuit without destroying any quantum advantage or requiring coherence times beyond those available on near-term devices.

What would settle it

Running the Mackey-Glass task with the quantum circuit replaced by an equivalent classical feature map and finding no reduction in mean squared error, or observing that two-qubit gate noise eliminates the accuracy gain, would falsify the claimed benefit.

Figures

Figures reproduced from arXiv: 2604.03469 by Aaron Z. Goldberg, Christoph Simon, Khabat Heshami, Utkarsh Singh.

**Figure 2.** Figure 2: FIG. 2: Schematic of proposed feedback-driven quantum reservoir computing model. At each timestep [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: FIG. 3: Model comparison on the Mackey–Glass dataset at [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: FIG. 4: Predicted vs. true signal for the quantum reservoir [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

**Figure 6.** Figure 6: FIG. 6: Impact of quantum circuit parameter [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗

**Figure 8.** Figure 8: FIG. 8: Performance (MSE score) of the quantum reservoir [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗

**Figure 9.** Figure 9: FIG. 9: Worst-case MSE for each noise type at the maximum [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗

**Figure 11.** Figure 11: FIG. 11: Dynamical analysis of the quantum reservoir. (a) [PITH_FULL_IMAGE:figures/full_fig_p010_11.png] view at source ↗

**Figure 12.** Figure 12: FIG. 12: MSE as a function of relaxation time [PITH_FULL_IMAGE:figures/full_fig_p011_12.png] view at source ↗

read the original abstract

Reservoir computing promises a fast method for handling large amounts of temporal data. This hinges on constructing a good reservoir--a dynamical system capable of transforming inputs into a high-dimensional representation while remembering properties of earlier data. In this work, we introduce a reservoir based on recurrent quantum feature maps where a fixed quantum circuit is reused to encode both current inputs and a classical feedback signal derived from previous outputs. We evaluate the model on the Mackey-Glass time-series prediction task using our recently introduced CP feature map, and find that it achieves lower mean squared error than standard classical baselines, including echo state networks and multilayer perceptrons, while maintaining compact circuit depth and qubit requirements. We further analyze memory capacity and show that the model effectively retains temporal information, consistent with its forecasting accuracy. Finally, we study the impact of realistic noise and find that performance is robust to several noise channels but remains sensitive to two-qubit gate errors, identifying a key limitation for near-term implementations.

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.

Desk Editor's Note private letter to a colleague

This paper adds classical recurrence to a quantum feature map reservoir and reports lower MSE than classical baselines on Mackey-Glass, though the noise results leave hardware feasibility open.

read the letter

The main takeaway is a recurrent quantum reservoir that reuses one fixed circuit by encoding both the current input and a classical feedback signal from prior outputs. On the Mackey-Glass task it records lower mean squared error than echo state networks and multilayer perceptrons while keeping circuit depth and qubit count modest. The construction itself is the new piece: prior quantum reservoir papers did not combine their CP feature map with this exact classical feedback loop. The memory capacity measurements line up with the forecasting accuracy, and the noise study is straightforward about robustness to several channels but clear sensitivity to two-qubit gate errors. That honesty about limitations is useful. The soft spots are the missing error bars, dataset splits, and statistical tests around the MSE numbers, plus the open question of whether the feedback encoding adds enough operations to exceed near-term coherence windows. The stress-test concern about error accumulation in the recurrent loop is worth checking in the methods, since the abstract already flags two-qubit errors as the main issue. If the full encoding stays light, the advantage could hold; otherwise the reported gain may shrink on hardware. This work is aimed at quantum machine learning researchers who want concrete hybrid constructions for time-series data. A reader already working on reservoir methods or near-term quantum models would get direct value from the map definition and the empirical setup. I would send it for peer review. The idea is implementable, the experiments are concrete, and the noise breakdown gives referees something specific to evaluate even if revisions on statistics and hardware modeling are needed.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes a recurrent quantum reservoir computing architecture based on fixed quantum circuits that encode both new inputs and a classical feedback signal derived from prior outputs via the CP feature map. It evaluates the model on the Mackey-Glass chaotic time-series prediction task, reporting lower mean squared error than classical echo state networks and multilayer perceptrons while using compact circuit depth and qubit count. Additional analyses cover memory capacity and robustness to several noise channels, with the key limitation identified as sensitivity to two-qubit gate errors.

Significance. If the empirical MSE advantage is confirmed with proper statistical controls, the recurrent construction offers a concrete route to near-term quantum advantage in temporal processing tasks by reusing a fixed circuit and avoiding deep variational optimization. The noise analysis usefully flags two-qubit gate errors as the dominant limitation, which is directly actionable for hardware implementations. The work is incremental rather than foundational, but the combination of recurrent feedback with a parameter-free feature map is a clear technical contribution.

major comments (3)

[Abstract / Results] Abstract and results section: the reported lower MSE versus ESN and MLP baselines is presented without error bars, cross-validation details, or statistical significance tests. This makes it impossible to determine whether the advantage is robust or could be explained by hyperparameter tuning differences.
[Methods / Recurrent construction] Recurrent construction (described in the methods): the classical feedback signal is re-encoded into the same fixed circuit, yet the manuscript provides no quantitative estimate of the additional state-preparation overhead, cumulative decoherence, or effective coherence time required per recurrent step. This directly affects whether the compact depth claim survives on near-term hardware with T1/T2 ~50-100 µs.
[Noise robustness analysis] Noise analysis: while sensitivity to two-qubit gate errors is noted, the paper does not simulate or bound the error accumulation rate across multiple recurrent steps, which is load-bearing for the claim that performance remains usable on NISQ devices.

minor comments (2)

[Methods] Notation for the CP feature map and the feedback encoding should be defined explicitly in a single location rather than referenced to prior work without restatement.
[Results] Figure captions for the Mackey-Glass predictions should include the exact train/test split sizes and the number of independent runs used to generate the plotted curves.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below and will incorporate revisions to improve statistical rigor, hardware feasibility estimates, and noise analysis.

read point-by-point responses

Referee: [Abstract / Results] Abstract and results section: the reported lower MSE versus ESN and MLP baselines is presented without error bars, cross-validation details, or statistical significance tests. This makes it impossible to determine whether the advantage is robust or could be explained by hyperparameter tuning differences.

Authors: We agree that the current results lack sufficient statistical controls. In the revised manuscript we will add error bars computed from multiple independent runs with different random seeds, provide explicit details on the cross-validation procedure used for hyperparameter selection, and include statistical significance tests (e.g., paired t-tests) comparing the recurrent quantum feature map against the ESN and MLP baselines. These additions will appear in the results section and, where space allows, in the abstract. revision: yes
Referee: [Methods / Recurrent construction] Recurrent construction (described in the methods): the classical feedback signal is re-encoded into the same fixed circuit, yet the manuscript provides no quantitative estimate of the additional state-preparation overhead, cumulative decoherence, or effective coherence time required per recurrent step. This directly affects whether the compact depth claim survives on near-term hardware with T1/T2 ~50-100 µs.

Authors: We acknowledge the absence of quantitative overhead estimates. The revised manuscript will include a new paragraph (or short subsection) in the methods that calculates the additional gate count and depth required to re-encode the classical feedback signal at each step, together with an estimate of cumulative decoherence under typical T1/T2 times of 50–100 µs. This will directly address the viability of the compact-depth claim on near-term hardware. revision: yes
Referee: [Noise robustness analysis] Noise analysis: while sensitivity to two-qubit gate errors is noted, the paper does not simulate or bound the error accumulation rate across multiple recurrent steps, which is load-bearing for the claim that performance remains usable on NISQ devices.

Authors: We agree that error accumulation over recurrent steps must be quantified. In the revision we will extend the noise section with explicit simulations of two-qubit gate error accumulation across increasing numbers of recurrent steps and will provide analytic bounds on the resulting degradation in prediction accuracy. These results will be presented alongside the existing single-step noise channels. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical performance claims rest on independent measurements

full rationale

The paper constructs a recurrent quantum feature map reservoir by reusing a fixed circuit to encode both new inputs and classical feedback from prior outputs, then reports measured mean squared error on the Mackey-Glass task that is lower than echo state networks and multilayer perceptrons. No derivation step equates a claimed prediction or uniqueness result to its own fitted inputs or prior self-citation by construction; the CP feature map reference is a modular building block whose definition is external to the present performance numbers. The central claims are therefore falsifiable via hardware runs or classical re-implementations and do not reduce to self-referential fitting or renaming.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that the CP feature map combined with classical feedback yields a sufficiently expressive reservoir; no free parameters or invented entities are explicitly introduced in the abstract.

axioms (1)

domain assumption The CP feature map provides a suitable high-dimensional encoding for both inputs and feedback signals
Referenced as the encoding method used in the recurrent circuit.

pith-pipeline@v0.9.0 · 5467 in / 1116 out tokens · 41956 ms · 2026-05-13T18:27:53.225118+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

a fixed quantum circuit is reused to encode both current inputs and a classical feedback signal derived from previous outputs... Ut = U†(Π(zt̃−1))U(xt)
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

CPMap... alternates between data-encoding rotation layers and carefully designed two-qubit entangling blocks

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

36 extracted references · 36 canonical work pages · 1 internal anchor

[1]

Jaeger and H

H. Jaeger and H. Haas, Harnessing nonlinearity: Predicting chaotic systems and saving energy in wireless communication, Science304, 78 (2004). 9

work page 2004
[2]

Schrauwen, D

B. Schrauwen, D. Verstraeten, and J. Van Campenhout, An overview of reservoir computing: theory, applications and im- plementations, inProceedings of the 15th European Symposium on Artificial Neural Networks(2007) pp. 471–482

work page 2007
[3]

Tanaka, T

G. Tanaka, T. Yamane, J. B. H´eroux, R. Nakane, N. Kanazawa, S. Takeda, H. Numata, D. Nakano, and A. Hirose, Recent ad- vances in physical reservoir computing: A review, Neural Net- works115, 100 (2019)

work page 2019
[4]

Maass, T

W. Maass, T. Natschl¨ager, and H. Markram, Real-time comput- ing without stable states: A new framework for neural com- putation based on perturbations, Neural computation14, 2531 (2002)

work page 2002
[6]

Verstraeten, B

D. Verstraeten, B. Schrauwen, M. D’Haene, and D. Stroobandt, An experimental unification of reservoir computing methods, Neural Networks20, 391 (2007), echo State Networks and Liq- uid State Machines

work page 2007
[7]

Fujii and K

K. Fujii and K. Nakajima, Harnessing disordered-ensemble quantum dynamics for machine learning, Physical Review Ap- plied8, 024030 (2017)

work page 2017
[8]

Negoro, K

M. Negoro, K. Mitarai, K. Fujii, K. Nakajima, and M. Kita- gawa, Machine learning with controllable quantum dynam- ics of a nuclear spin ensemble in a solid, arXiv preprint arXiv:1806.10910 (2018)

work page arXiv 2018
[9]

J. e. a. Nokkala, Gaussian states of continuous-variable quan- tum systems provide universal and versatile reservoir comput- ing, Communications Physics4, 53 (2021)

work page 2021
[10]

J.-D. Chen, H. I. Nurdin, and N. Yamamoto, Temporal infor- mation processing on noisy quantum computers via reservoir computing, Physical Review Applied14, 024065 (2020)

work page 2020
[11]

Nurdin, and Naoki Yamamoto

H. Yasuda and N. Yamamoto, Quantum reservoir computing with repeated measurements on superconducting devices, arXiv preprint arXiv:2310.06706 (2023)

work page arXiv 2023
[12]

arXiv preprint arXiv:2407.02553 , year=

M. Kornja ˇca, 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. Hu- ber, 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
[13]

A. e. a. Sannia, Dissipation as a resource for quantum reservoir computing, Quantum8, 1291 (2024)

work page 2024
[14]

L. C. Goviaet al., Quantum reservoir computing with a sin- gle nonlinear oscillator, Physical Review Research3, 013077 (2021)

work page 2021
[15]

Mujal, J

P. Mujal, J. Guerrero, D. Garcia-Beni, G. F. Calvo, and A. Alarcon, Time-series quantum reservoir computing with weak and projective measurements, npj Quantum Information 9, 16 (2023)

work page 2023
[16]

Murauer, A

T. Murauer, A. Matsuura, and Y . Yamamoto, Feedback connec- tions in quantum reservoir computing with mid-circuit mea- surements, arXiv preprint (2025), arXiv:2503.22380 [quant- ph]

work page arXiv 2025
[17]

Kobayashi, K

K. Kobayashi, K. Fujii, and Y . Yamamoto, Feedback-driven quantum reservoir computing for time-series analysis, PRX Quantum5, 040325 (2024)

work page 2024
[18]

Kobayashi, K

K. Kobayashi, K. Fujii, and N. Yamamoto, Feedback-driven quantum reservoir computing for time-series analysis, PRX Quantum5, 040325 (2024)

work page 2024
[19]

Feedback-driven recurrent quantum neural network universality, 2026

L. Gonon, R. Mart ´ınez-Pe˜na, and J.-P. Ortega, Feedback- driven recurrent quantum neural network universality (2026), arXiv:2506.16332 [quant-ph]

work page arXiv 2026
[20]

Feedback-enhanced quantum reservoir com- puting with weak measurements, 2025

R. Monomi, Y . Yamamoto, and K. Fujii, Feedback-enhanced quantum reservoir computing with weak measurements, arXiv preprint (2025), arXiv:2503.17939 [quant-ph]

work page arXiv 2025
[21]

M. e. a. Domingo, Taking advantage of noise in quantum reser- voir computing, Scientific Reports13, 11591 (2023)

work page 2023
[22]

L. e. a. ˇCindrak, Enhancing the performance of quantum reser- voir computing and solving the time-complexity problem by artificial memory restriction, Physical Review Research6, 013076 (2024)

work page 2024
[23]

H. Goto, M. C. Tran, and K. Nakajima, Universal approxima- tion property of quantum machine learning models in quantum- enhanced feature spaces, Physical Review Letters127, 090506 (2021)

work page 2021
[24]

Iterative Quantum Feature Maps

K. Matsumoto and M. C. Tran, Iterative quantum feature maps, arXiv preprint (2025), arXiv:2506.19461 [quant-ph]

work page internal anchor Pith review Pith/arXiv arXiv 2025
[25]

Jaeger,The ”echo state” approach to analysing and train- ing recurrent neural networks-with an erratum note, Tech

H. Jaeger,The ”echo state” approach to analysing and train- ing recurrent neural networks-with an erratum note, Tech. Rep. (Bonn, Germany: German National Research Center for Infor- mation Technology GMD Technical Report, 2001)

work page 2001
[26]

Luko ˇseviˇcius and H

M. Luko ˇseviˇcius and H. Jaeger, Reservoir computing ap- proaches to recurrent neural network training, Computer Sci- ence Review3, 127 (2009)

work page 2009
[27]

I. B. Yildiz, H. Jaeger, and S. J. Kiebel, Re-visiting the echo state property, Neural networks35, 1 (2012)

work page 2012
[28]

Boyd and L

S. Boyd and L. O. Chua,Fading memory and the problem of approximating nonlinear operators with Volterra series(IEEE Transactions on circuits and systems, 1985)

work page 1985
[29]

Grigoryeva and J.-P

L. Grigoryeva and J.-P. Ortega, Echo state networks are univer- sal, Neural Networks108, 495 (2018)

work page 2018
[30]

Gonc ¸alves, N

L. Gonc ¸alves, N. Marques, R. Marques, and R. Rego, Reser- voir computing universality with stochastic inputs, IEEE Trans- actions on Neural Networks and Learning Systems31, 4749 (2020)

work page 2020
[31]

Jaeger,Short term memory in echo state networks, Tech

H. Jaeger,Short term memory in echo state networks, Tech. Rep. (GMD-Forschungszentrum Informationstechnik, 2002)

work page 2002
[32]

M. C. Mackey and L. Glass, Oscillation and chaos in physio- logical control systems, Science197, 287 (1977)

work page 1977
[33]

Singh, J.-F

U. Singh, J.-F. Laprade, A. Z. Goldberg, and K. Heshami, A resource efficient quantum kernel (2025), arXiv:2507.03689 [quant-ph]

work page arXiv 2025
[34]

Havl´ıˇcek, A

V . Havl´ıˇcek, A. D. C´orcoles, K. Temme, A. W. Harrow, A. Kan- dala, J. M. Chow, and J. M. Gambetta, Supervised learning with quantum-enhanced feature spaces, Nature567, 209 (2019)

work page 2019
[35]

Qiskit contributors, Qiskit: An open-source framework for quantum computing (2023)

work page 2023
[36]

Asfaw, L

A. Asfaw, L. Bello, Y . Ben-Haim, S. Bravyi, L. Capelluto, A. C. Vazquez, J. Ceroni, R. Chen, A. Frisch, J. Gambetta, S. Gar- ion, L. Gil, S. D. L. P. Gonzalez, F. Harkins, T. Imamichi, D. McKay, A. Mezzacapo, Z. Minev, R. Movassagh, G. Nan- nicini, P. Nation, A. Phan, M. Pistoia, A. Rattew, J. Schaefer, J. Shabani, J. Smolin, K. Temme, M. Tod, and S. Woo...

work page 2020
[37]

Dambre, D

J. Dambre, D. Verstraeten, B. Schrauwen, and S. Massar, In- formation processing capacity of dynamical systems, Scientific reports2, 514 (2012). 10 A. Effect of Feedback Strength and Readout Truncation We next examine how two core parameters—feedback strengthαand readout limitλ—influence forecasting perfor- mance. As described previously,αcontrols the con...

work page 2012

[1] [1]

Jaeger and H

H. Jaeger and H. Haas, Harnessing nonlinearity: Predicting chaotic systems and saving energy in wireless communication, Science304, 78 (2004). 9

work page 2004

[2] [2]

Schrauwen, D

B. Schrauwen, D. Verstraeten, and J. Van Campenhout, An overview of reservoir computing: theory, applications and im- plementations, inProceedings of the 15th European Symposium on Artificial Neural Networks(2007) pp. 471–482

work page 2007

[3] [3]

Tanaka, T

G. Tanaka, T. Yamane, J. B. H´eroux, R. Nakane, N. Kanazawa, S. Takeda, H. Numata, D. Nakano, and A. Hirose, Recent ad- vances in physical reservoir computing: A review, Neural Net- works115, 100 (2019)

work page 2019

[4] [4]

Maass, T

W. Maass, T. Natschl¨ager, and H. Markram, Real-time comput- ing without stable states: A new framework for neural com- putation based on perturbations, Neural computation14, 2531 (2002)

work page 2002

[5] [6]

Verstraeten, B

D. Verstraeten, B. Schrauwen, M. D’Haene, and D. Stroobandt, An experimental unification of reservoir computing methods, Neural Networks20, 391 (2007), echo State Networks and Liq- uid State Machines

work page 2007

[6] [7]

Fujii and K

K. Fujii and K. Nakajima, Harnessing disordered-ensemble quantum dynamics for machine learning, Physical Review Ap- plied8, 024030 (2017)

work page 2017

[7] [8]

Negoro, K

M. Negoro, K. Mitarai, K. Fujii, K. Nakajima, and M. Kita- gawa, Machine learning with controllable quantum dynam- ics of a nuclear spin ensemble in a solid, arXiv preprint arXiv:1806.10910 (2018)

work page arXiv 2018

[8] [9]

J. e. a. Nokkala, Gaussian states of continuous-variable quan- tum systems provide universal and versatile reservoir comput- ing, Communications Physics4, 53 (2021)

work page 2021

[9] [10]

J.-D. Chen, H. I. Nurdin, and N. Yamamoto, Temporal infor- mation processing on noisy quantum computers via reservoir computing, Physical Review Applied14, 024065 (2020)

work page 2020

[10] [11]

Nurdin, and Naoki Yamamoto

H. Yasuda and N. Yamamoto, Quantum reservoir computing with repeated measurements on superconducting devices, arXiv preprint arXiv:2310.06706 (2023)

work page arXiv 2023

[11] [12]

arXiv preprint arXiv:2407.02553 , year=

M. Kornja ˇca, 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. Hu- ber, 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

[12] [13]

A. e. a. Sannia, Dissipation as a resource for quantum reservoir computing, Quantum8, 1291 (2024)

work page 2024

[13] [14]

L. C. Goviaet al., Quantum reservoir computing with a sin- gle nonlinear oscillator, Physical Review Research3, 013077 (2021)

work page 2021

[14] [15]

Mujal, J

P. Mujal, J. Guerrero, D. Garcia-Beni, G. F. Calvo, and A. Alarcon, Time-series quantum reservoir computing with weak and projective measurements, npj Quantum Information 9, 16 (2023)

work page 2023

[15] [16]

Murauer, A

T. Murauer, A. Matsuura, and Y . Yamamoto, Feedback connec- tions in quantum reservoir computing with mid-circuit mea- surements, arXiv preprint (2025), arXiv:2503.22380 [quant- ph]

work page arXiv 2025

[16] [17]

Kobayashi, K

K. Kobayashi, K. Fujii, and Y . Yamamoto, Feedback-driven quantum reservoir computing for time-series analysis, PRX Quantum5, 040325 (2024)

work page 2024

[17] [18]

Kobayashi, K

K. Kobayashi, K. Fujii, and N. Yamamoto, Feedback-driven quantum reservoir computing for time-series analysis, PRX Quantum5, 040325 (2024)

work page 2024

[18] [19]

Feedback-driven recurrent quantum neural network universality, 2026

L. Gonon, R. Mart ´ınez-Pe˜na, and J.-P. Ortega, Feedback- driven recurrent quantum neural network universality (2026), arXiv:2506.16332 [quant-ph]

work page arXiv 2026

[19] [20]

Feedback-enhanced quantum reservoir com- puting with weak measurements, 2025

R. Monomi, Y . Yamamoto, and K. Fujii, Feedback-enhanced quantum reservoir computing with weak measurements, arXiv preprint (2025), arXiv:2503.17939 [quant-ph]

work page arXiv 2025

[20] [21]

M. e. a. Domingo, Taking advantage of noise in quantum reser- voir computing, Scientific Reports13, 11591 (2023)

work page 2023

[21] [22]

L. e. a. ˇCindrak, Enhancing the performance of quantum reser- voir computing and solving the time-complexity problem by artificial memory restriction, Physical Review Research6, 013076 (2024)

work page 2024

[22] [23]

H. Goto, M. C. Tran, and K. Nakajima, Universal approxima- tion property of quantum machine learning models in quantum- enhanced feature spaces, Physical Review Letters127, 090506 (2021)

work page 2021

[23] [24]

Iterative Quantum Feature Maps

K. Matsumoto and M. C. Tran, Iterative quantum feature maps, arXiv preprint (2025), arXiv:2506.19461 [quant-ph]

work page internal anchor Pith review Pith/arXiv arXiv 2025

[24] [25]

Jaeger,The ”echo state” approach to analysing and train- ing recurrent neural networks-with an erratum note, Tech

H. Jaeger,The ”echo state” approach to analysing and train- ing recurrent neural networks-with an erratum note, Tech. Rep. (Bonn, Germany: German National Research Center for Infor- mation Technology GMD Technical Report, 2001)

work page 2001

[25] [26]

Luko ˇseviˇcius and H

M. Luko ˇseviˇcius and H. Jaeger, Reservoir computing ap- proaches to recurrent neural network training, Computer Sci- ence Review3, 127 (2009)

work page 2009

[26] [27]

I. B. Yildiz, H. Jaeger, and S. J. Kiebel, Re-visiting the echo state property, Neural networks35, 1 (2012)

work page 2012

[27] [28]

Boyd and L

S. Boyd and L. O. Chua,Fading memory and the problem of approximating nonlinear operators with Volterra series(IEEE Transactions on circuits and systems, 1985)

work page 1985

[28] [29]

Grigoryeva and J.-P

L. Grigoryeva and J.-P. Ortega, Echo state networks are univer- sal, Neural Networks108, 495 (2018)

work page 2018

[29] [30]

Gonc ¸alves, N

L. Gonc ¸alves, N. Marques, R. Marques, and R. Rego, Reser- voir computing universality with stochastic inputs, IEEE Trans- actions on Neural Networks and Learning Systems31, 4749 (2020)

work page 2020

[30] [31]

Jaeger,Short term memory in echo state networks, Tech

H. Jaeger,Short term memory in echo state networks, Tech. Rep. (GMD-Forschungszentrum Informationstechnik, 2002)

work page 2002

[31] [32]

M. C. Mackey and L. Glass, Oscillation and chaos in physio- logical control systems, Science197, 287 (1977)

work page 1977

[32] [33]

Singh, J.-F

U. Singh, J.-F. Laprade, A. Z. Goldberg, and K. Heshami, A resource efficient quantum kernel (2025), arXiv:2507.03689 [quant-ph]

work page arXiv 2025

[33] [34]

Havl´ıˇcek, A

V . Havl´ıˇcek, A. D. C´orcoles, K. Temme, A. W. Harrow, A. Kan- dala, J. M. Chow, and J. M. Gambetta, Supervised learning with quantum-enhanced feature spaces, Nature567, 209 (2019)

work page 2019

[34] [35]

Qiskit contributors, Qiskit: An open-source framework for quantum computing (2023)

work page 2023

[35] [36]

Asfaw, L

A. Asfaw, L. Bello, Y . Ben-Haim, S. Bravyi, L. Capelluto, A. C. Vazquez, J. Ceroni, R. Chen, A. Frisch, J. Gambetta, S. Gar- ion, L. Gil, S. D. L. P. Gonzalez, F. Harkins, T. Imamichi, D. McKay, A. Mezzacapo, Z. Minev, R. Movassagh, G. Nan- nicini, P. Nation, A. Phan, M. Pistoia, A. Rattew, J. Schaefer, J. Shabani, J. Smolin, K. Temme, M. Tod, and S. Woo...

work page 2020

[36] [37]

Dambre, D

J. Dambre, D. Verstraeten, B. Schrauwen, and S. Massar, In- formation processing capacity of dynamical systems, Scientific reports2, 514 (2012). 10 A. Effect of Feedback Strength and Readout Truncation We next examine how two core parameters—feedback strengthαand readout limitλ—influence forecasting perfor- mance. As described previously,αcontrols the con...

work page 2012