Pith. sign in

REVIEW 3 major objections 4 minor 135 references

Rethinking Reservoir Pruning: A Dynamical Perspective for Echo State Networks

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read The paper proposes a one-shot, data-driven criterion that ranks echo-state reservoir neurons by their contribution to dominant trajectory-dependent transition modes, and shows that removing the lowest-ranked 20% preserves or improves…

desk verdict DMP's Gramian-based pruning is a new and mostly honest incremental contribution, but the one-shot assumption and the over-promoted dynamical edge need scrutiny before publication. read the letter →

arxiv 2608.04593 v1 pith:RL42Y5NV submitted 2026-08-05 cs.LG cs.AImath.DS

classification cs.LGcs.AImath.DS
keywords echostatenetworksreservoirpruningJacobianGramiandynamicalmodestime-seriesforecastingstructuredchaoticsystemsmodelcompression
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper proposes Dynamical Mode Pruning (DMP), a one-shot method for compressing Echo State Networks by removing reservoir neurons that contribute least to the input-driven transition structure. DMP builds a trajectory-averaged Jacobian Gramian from the driven reservoir, keeps the dominant eigendirections, and scores each neuron by how strongly its coordinate participates in those directions. Pruned networks retrain only the linear readout, leaving recurrent weights fixed. In tests on chaotic Mackey-Glass and four real-world time series, pruning from 1,000 to 800 neurons lowers mean NRMSE relative to the full ESN on every dataset, cuts storage by 35.7%, and speeds up inference by 1.90 to 1.97 times. The paper's claim is that dynamical influence is a useful and principled criterion for reservoir refinement beyond static connectivity or activation statistics.

What carries the argument

The object that carries the argument is the trajectory-averaged Jacobian Gramian $G = \frac{1}{T}\sum_{t=1}^{T} J(t)^{\top}J(t)$, with $J(t) = (1-a)I + a\,\mathrm{diag}(f'(z(t)))W$ for the leaky ESN transition. Its eigendecomposition $G = V\Lambda V^{\top}$ gives directions in state space along which input-driven local perturbations grow most strongly on average; after normalizing eigenvalues and keeping the smallest number of modes whose cumulative energy reaches $\tau = 0.9$, each neuron's score is $\eta_i = \sum_{k=1}^{r}\tilde{\lambda}_k V_{ik}^{2}$, the mode-energy-weighted participation of coordinate $i$ in the dominant subspace. Low-$\eta_i$ neurons are removed from $W$, $W^{\mathrm{in}}$, and $W^{\mathrm{fb}}$, and only the readout is retrained by ridge regression. This turns pruning into a data-dependent dynamical-subspace calculation rather than a static graph or activation heuristic.

What would settle it

A direct check: after pruning at 30%, recompute the trajectory-averaged Jacobian Gramian for the pruned reservoir and compare its dominant subspace with the original; if the principal-angle overlap is small, and re-ranking with the new Gramian changes the retained set and improves NRMSE, then the one-shot assumption fails.

Watch

Extended reading notes

Core claim

On the paper's own terms, DMP establishes that a random untrained ESN reservoir contains dynamical redundancy that can be identified from the trajectory it actually follows. The importance of a neuron is defined as its weighted participation in the leading modes of the trajectory-averaged Jacobian Gramian, truncated to modes carrying at least 90% of the Gramian energy. Removing the lowest-scoring neurons in one shot and refitting only the readout improves or preserves mean NRMSE compared with the full reservoir, with the clearest gains on Mackey-Glass and Electricity, and remains competitive with newly initialized smaller ESNs and random pruning under a matched ten-seed protocol. Spectral-radius rescaling after pruning is not needed; the unscaled pruned reservoir is the recommended configuration.

Load-bearing premise

The load-bearing premise is that the dynamical-mode ranking computed once from the unpruned reservoir remains valid for the pruned reservoir; if pruning shifts the dominant transition directions, the neuron scores become miscalibrated.

Editorial extensions

If this is right

  • At 20% pruning (1000 to 800 neurons), DMP yields lower mean NRMSE than the full ESN on all five datasets in the matched ten-seed protocol, with the largest separation on Mackey-Glass.
  • The pruned model uses 640,000 instead of 1,000,000 recurrent parameters, is 35.7% smaller on disk, and runs at 1.90 to 1.97 times inference speedup.
  • Matched random pruning also beats the full ESN on most datasets, so part of the benefit is simply smaller reservoir size; DMP's criterion adds a smaller, dataset-dependent margin.
  • Pruning at 10% removes too little redundancy and 30% often degrades accuracy, so the one-shot Gramian ranking is reliable only for moderate pruning ratios.
  • Spectral-radius rescaling after pruning does not help; DMP without rescaling is the main configuration.

Reading between the lines

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

  • A natural extension the paper does not test is iterative DMP: recompute the Gramian on the pruned reservoir and prune again; this would directly test the one-shot approximation and may recover the 30% pruning regime.
  • The same Gramian-energy criterion could also select reservoir capacity per task, since the cumulative-energy elbow provides a data-driven bound on how many neurons the dynamics actually require.
  • Because random pruning already captures much of the gain in real-world datasets, a practical deployment might treat DMP as a tie-breaker among several cheap criteria, with its largest added value on dynamics-dominated tasks such as chaotic forecasting.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes Dynamical Mode Pruning (DMP), a one-shot pruning method for Echo State Networks that ranks reservoir neurons by their participation in dominant modes of a trajectory-averaged Jacobian Gramian. The method computes J(t) = (1-a)I + aD(t)W along a teacher-forced trajectory, forms G = (1/T)∑J(t)^T J(t), retains the top-r eigenmodes up to an energy threshold tau, and assigns each neuron a score from the weighted squared eigenvector components. The lowest-scoring neurons are removed, and only the readout is refitted; an optional spectral-radius rescaling is treated as an ablation. The paper reports benchmark comparisons on Mackey–Glass and four real-world time series, including a controlled matched evaluation at N=1000→800 over ten paired seeds, where DMP has the lowest mean NRMSE among the compared methods and yields a 35.7% model-size reduction with roughly 1.9–2.0x inference speedup. The limitations section acknowledges the offline cost and the fact that the original Gramian no longer exactly describes the reduced reservoir after pruning.

Significance. If the central claim holds, DMP would offer a principled, data-dependent alternative to static connectivity or activation-based pruning for ESNs, with a one-shot criterion and negligible inference-time overhead. The paper has notable strengths: the Jacobian derivation in Eq. (4) is correct for the leaky ESN, the controlled matched evaluation pairs DMP and random pruning on the same reservoir and data split, the code is publicly available, and the limitations section is unusually candid about the one-shot approximation and the 30%-pruning degradation. However, the empirical evidence that the dynamical score beats matched random pruning or simple static criteria is currently weak on several real-world datasets, and the load-bearing one-shot Gramian assumption is acknowledged but not directly tested. The significance of the paper therefore depends on additional validation rather than on the theoretical construction alone.

major comments (3)
  1. [Table S2 / Section 4] The controlled matched evaluation does not support a robust advantage of DMP over matched random pruning on four of the five datasets. Table S2 reports mean DMP-minus-random NRMSE differences of -0.0026 (Electricity, 6/10 wins), -0.0102 (Temperature, 6/10), -0.0001 (Solar, 5/10), and -0.0001 (Wind, 5/10); only Mackey–Glass shows a clear separation with 9/10 wins. Because the paper's stated contribution is that trajectory-dependent dynamical influence is more informative than an arbitrary pruning mask, this evidence is load-bearing. Please report paired confidence intervals and win/loss statistics for the DMP-versus-random comparison, and either soften the claim to 'preserves' for Solar and Wind or provide additional evidence for improvement on those datasets.
  2. [Section 5.1 / Algorithm 1] The one-shot Gramian assumption is acknowledged in Section 5.1 but is not directly validated. After pruning, the reduced Jacobian is J_I(t) = (1-a)I + aD_I(t)W[I,I], which is not the principal submatrix of the original J(t) because the retained pre-activations lose the term W[I,I^c]x_{I^c}(t). The paper states that the original Gramian 'no longer exactly describes the reduced system' but provides no evidence that the dominant modes of the original Gramian remain dominant for the reduced reservoir. Please add an experiment that measures, at each pruning ratio, the subspace distance between the top-r eigenspace of G and the Gramian of the reduced reservoir, or the rank correlation between neuron scores computed before and after pruning. This would directly test the central assumption and would also help interpret the degradation observed at 30% pruning.
  3. [Table S1 / Section A.1] The expanded comparison with static and simple baselines weakens the claim that DMP outperforms other pruning criteria. Under the uniformly rescaled protocol in Table S1, DMP+ρ has the lowest mean NRMSE only on Temperature; activation-variance+ρ is best on Mackey–Glass, magnitude+ρ is best on Electricity and Wind, and Random+ρ is best on Solar. Because the main unscaled DMP configuration is not included in this table, the comparison is not a complete matched test of DMP against static criteria. Please either include the unscaled DMP (and unscaled baseline variants) in the matched protocol with paired statistics, or explicitly restrict the paper's conclusion to 'DMP improves or preserves accuracy relative to the full ESN' rather than claiming superiority over static structural importance measures.
minor comments (4)
  1. [Equations (11)–(13)] The notation 'eλk' is not defined and is difficult to read; please use a standard symbol such as λ̂_k and state explicitly that it is the eigenvalue normalized by the trace of G.
  2. [Table 2] For Wind, the Base ESN mean is 19.80 with a standard deviation of 42.69 and a confidence interval reported as [0.0000, 72.8170], so the headline 89.8% mean reduction is driven by a small number of seeds. Consider reporting geometric means, log-scaled confidence intervals, or per-seed values to make the variability transparent.
  3. [Table 1 / Section 5] The caption of Table 1 explains that DMP runtime and memory include the dense Jacobian–Gramian pipeline and are not comparable to Table 4, but the main text should repeat this caveat at the first mention of the 'Runtime' and 'Peak Offline Memory' columns so that readers do not interpret DMP's 1836 MB peak usage as the cost of the final method.
  4. [Section 5.1] The sentence suggesting that a 'large change in spectral radius after pruning' or a 'strong shift in the Gramian eigenvalue spectrum' can indicate one-shot unreliability is not operationalized; please specify a concrete diagnostic procedure or threshold.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: DMP's Gramian-based scores are computed from reservoir dynamics and evaluated on held-out test data; the one-shot approximation is a stated limitation, not a definitional loop.

full rationale

The derivation chain is self-contained. Neuron importance in DMP is defined by Eq. (13) as the energy-weighted participation of each coordinate in the dominant eigenvectors of the trajectory-averaged Jacobian Gramian (Eq. (7)), and pruning follows by taking the top-N' indices (Eq. (14)). The readout is then refitted by ridge regression on the reduced reservoir, and all reported gains are measured on held-out test data. No equation in the paper reduces to a fitted value, no fitted parameter is renamed as a prediction, and the method does not invoke a uniqueness theorem or an ansatz smuggled in via citation. The only self-citation, [Laudari, 2026], appears in Related Work as one example of centrality-based reservoir pruning and is not load-bearing for the DMP construction. The paper explicitly acknowledges in Section 5.1 that after pruning 'the Gramian computed from the original reservoir no longer exactly describes the reduced system'; this is an honest stated limitation about the one-shot approximation, and a correctness or robustness concern rather than circularity. Test data are not used to set the Gramian threshold, pruning ratio, or readout regularization, so the empirical claim that DMP improves or preserves forecasting accuracy remains an independent experimental result.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The core heuristic is the identity between dominant Gramian modes and task-critical dynamics. No physical entities are introduced. The free parameters balance the method's behavior; none are fit to test labels, but tau and the pruning ratio shape the reported gains.

free parameters (5)
  • energy threshold tau = 0.9
    Sets the number of dominant modes r via Eq (12); fixed from an elbow heuristic on the cumulative Gramian energy across datasets, but it directly controls which modes contribute to neuron scores.
  • pruning ratio (target size N') = 20% (N'=800 from N=1000 in main results)
    Determines how many lowest-scoring neurons are removed; the paper reports 10/20/30% sensitivity but uses 20% as the fixed main setting, chosen a priori.
  • leak rate a = 0.5
    Enters the state update and the Jacobian in Eq (4); fixed across datasets, not tuned, but it changes the Gramian and therefore the pruning ranking.
  • ridge coefficient beta = 1e-4
    Regularization for the readout ridge regression; standard fixed ESN hyperparameter.
  • spectral radius target rho_target (DMP+rho ablation) = unspecified
    Used only in the ablation variant of Eq (18) to rescale the pruned reservoir; the target value is not reported in the main text or supplement.
assumptions (5)
  • domain assumption Spectral radius constraint rho(W) < 1 is sufficient for the Echo State Property in the tested regimes
    Invoked in Section 3: 'the spectral radius is constrained as rho(W)<1, which is a standard practical condition associated with the Echo State Property.' The exact ESP also depends on leak rate, input scaling, and nonlinearity, so this is an assumption about stability for the experimental configurations.
  • domain assumption Teacher forcing: y(t) is independent of x(t) when computing the Jacobian
    Stated in Section 3.1: 'Under this setting, y(t) is supplied externally and is therefore treated as independent of x(t) when computing the derivative.' If the ESN were used closed-loop with feedback, the Jacobian would include an extra feedback term, and the Gramian-based scores would change.
  • domain assumption Dominant eigendirections of the trajectory-averaged Jacobian Gramian capture the transition structure relevant for forecasting
    Core heuristic of DMP (Section 3.1): the quadratic form in Eq (8) is interpreted as average perturbation energy, and the leading r modes are assumed to represent the important dynamics. This is a modeling choice, not a proven equivalence.
  • domain assumption One-shot approximation: the Gramian of the original reservoir remains a valid importance measure after pruning
    Acknowledged in Section 5.1: 'the Gramian computed from the original reservoir no longer exactly describes the reduced system.' The method does not iterate or re-estimate the Gramian, so the validity of the ranking post-pruning is assumed.
  • standard math Standard linear algebra: G is symmetric positive semidefinite and admits a nonnegative eigendecomposition
    Used without proof in Section 3.1; each J(t)^T J(t) is PSD, so the average G is PSD and the eigendecomposition with real nonnegative eigenvalues is guaranteed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Rethinking Reservoir Pruning: A Dynamical Perspective for Echo State Networks." pith.science (2026). https://pith.science/paper/RL42Y5NV

@misc{pith2026260804593,
  author       = {Pith},
  title        = {Pith review of: Rethinking Reservoir Pruning: A Dynamical Perspective for Echo State Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RL42Y5NV}},
  note         = {Machine review of arXiv:2608.04593}
}
read the original abstract

Echo State Networks (ESNs) offer an efficient framework for temporal prediction, but their randomly initialized reservoirs are often over-parameterized and dynamically redundant. Existing pruning methods largely rely on static connectivity or activation statistics, which may overlook neurons that shape input-driven state transitions. We propose Dynamical Mode Pruning (DMP), a reservoir pruning method that ranks neurons by their contribution to dominant transition modes obtained from a trajectory-averaged Jacobian Gramian. DMP removes low-impact units and retrains only the readout. Experiments on chaotic and real-world time-series benchmarks show that DMP improves or preserves forecasting accuracy while reducing redundant reservoir components. Our results suggest that dynamical influence is a useful criterion for reservoir refinement beyond static structural importance alone.

Figures

Figures reproduced from arXiv: 2608.04593 by the authors.

Figure 1
Figure 1. ESN architecture with input weights Win, recurrent reservoir W, and readout Wout. Optional feedback Wf b enables output-driven dynamics. The reservoir is initialized to support stable input-driven state evolution. In this work, we introduce Dynamical Mode Pruning (DMP), a one-shot structured prun￾ing method that measures neuron importance through trajectory-dependent reservoir dynam￾ics. DMP constructs a trajectory-… view at source ↗
Figure 2
Figure 2. DMP framework. Reservoir states are collected from the driven ESN, and trajectory [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. DMP analysis: (a) Ground-truth and predicted time series after pruning. (b) Cumulative energy of the trajectory-averaged Jacobian Gramian(τ = 0.9) and the corre￾sponding mode rank indicated. (c) DMP neuron-importance scores, distinguishing re￾tained and pruned reservoir neurons. (d) Dis￾tribution of importance scores for the retained and pruned neuron groups. Neuron index (j) Neuron index (j) Neuron index (i) Reserv… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Representative structural analysis of DMP. Top: recurrent reservoir weight matri￾ces before and after pruning. Bottom: eigen￾value spectrum of the trajectory-averaged Ja￾cobian Gramian used to identify the dominant dynamical subspace. lower-ranked components. Matched a…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

135 extracted references · 77 canonical work pages

  1. [1]

    Climate data online

    Australian Bureau of Meteorology . Climate data online. https://www.bom.gov.au/climate/data/, 2026. Accessed: 2026-03-31

  2. [2]

    Residential electricity consumption data

    Australian Government Data Portal . Residential electricity consumption data. https://data.gov.au/, 2024. Accessed: 2026-03-31

  3. [3]

    Investigating Echo State Networks dynamics by means of recurrence analysis

    Filippo Maria Bianchi, Lorenzo Livi, and Cesare Alippi. Investigating Echo State Networks dynamics by means of recurrence analysis. IEEE Transactions on Neural Networks and Learning Systems , 29 0 (2): 0 427--439, 2016

  4. [4]

    What is the state of neural network pruning? Proceedings of Machine Learning and Systems, 2: 0 129--146, 2020

    Davis Blalock, Jose Javier Gonzalez Ortiz, Jonathan Frankle, and John Guttag. What is the state of neural network pruning? Proceedings of Machine Learning and Systems, 2: 0 129--146, 2020. URL https://proceedings.mlsys.org/paper_files/paper/2020/hash/6c44dc73014d66ba49b28d483a8f8b0d-Abstract.html

  5. [5]

    Centrality measures in networks

    Francis Bloch, Matthew O Jackson, and Pietro Tebaldi. Centrality measures in networks. Social Choice and Welfare, 61 0 (2): 0 413--453, 2023

  6. [6]

    On the sensitivity of centrality metrics

    Lucia Cavallaro, Pasquale De Meo, Giacomo Fiumara, and Antonio Liotta. On the sensitivity of centrality metrics. PLOS ONE, 19 0 (5): 0 e0299255, 2024

  7. [7]

    Recurrent neural network pruning using dynamical systems and iterative fine-tuning

    Christos Chatzikonstantinou, Dimitrios Konstantinidis, Kosmas Dimitropoulos, and Petros Daras. Recurrent neural network pruning using dynamical systems and iterative fine-tuning. Neural Networks, 143: 0 475--488, 2021

  8. [8]

    Pruning and regularization in reservoir computing

    Xavier Dutoit, Benjamin Schrauwen, et al. Pruning and regularization in reservoir computing. Neurocomputing, 72 0 (7--9): 0 1534--1546, 2009

Show all 135 references
  1. [9]

    Data-efficient structured pruning via submodular optimization

    Marwa El Halabi, Suraj Srinivas, and Simon Lacoste-Julien. Data-efficient structured pruning via submodular optimization. In Advances in Neural Information Processing Systems, volume 35, pages 36613--36626. Curran Associates, Inc., 2022. URL https://proceedings.neurips.cc/pape...

  2. [10]

    The lottery ticket hypothesis: Finding sparse, trainable neural networks

    Jonathan Frankle and Michael Carbin. The lottery ticket hypothesis: Finding sparse, trainable neural networks. In International Conference on Learning Representations, 2019. URL https://iclr.cc/virtual/2019/oral/1156

  3. [11]

    SPDY : Accurate pruning with speedup guarantees

    Elias Frantar and Dan Alistarh. SPDY : Accurate pruning with speedup guarantees. In Proceedings of the 39th International Conference on Machine Learning, volume 162 of Proceedings of Machine Learning Research, pages 6726--6743. PMLR, 2022. URL https://proceedings.mlr.press/v16...

  4. [12]

    Centrality in social networks: Conceptual clarification

    Linton C Freeman. Centrality in social networks: Conceptual clarification. Social Networks, 1 0 (3): 0 215--239, 1979

  5. [14]

    Local Lyapunov exponents of deep Echo State Networks

    Claudio Gallicchio, Alessio Micheli, and Luca Silvestri. Local Lyapunov exponents of deep Echo State Networks . Neurocomputing, 298: 0 34--45, 2018

  6. [17]

    Song Han, Jeff Pool, John Tran, and William J. Dally. Learning both weights and connections for efficient neural network. In Advances in Neural Information Processing Systems, volume 28. Curran Associates, Inc., 2015. URL https://proceedings.neurips.cc/paper/2015/hash/ae0eb3ee...

  7. [18]

    Sparse double descent: Where network pruning aggravates overfitting

    Zheng He, Zeke Xie, Quanzhi Zhu, and Zengchang Qin. Sparse double descent: Where network pruning aggravates overfitting. In Proceedings of the 39th International Conference on Machine Learning, volume 162 of Proceedings of Machine Learning Research, pages 8635--8659. PMLR, 202...

  8. [19]

    Semi-supervised Echo State Network with partial correlation pruning for time-series variables prediction in industrial processes

    Jian Huang, Fan Wang, Xu Yang, and Qing Li. Semi-supervised Echo State Network with partial correlation pruning for time-series variables prediction in industrial processes. Measurement Science and Technology, 34 0 (9): 0 095106, 2023

  9. [20]

    The ``Echo State'' approach to analysing and training recurrent neural networks---with an erratum note

    Herbert Jaeger. The ``Echo State'' approach to analysing and training recurrent neural networks---with an erratum note. GMD Technical Report , 148 0 (34): 0 13, 2001

  10. [21]

    Optimization and applications of Echo State Networks with leaky-integrator neurons

    Herbert Jaeger, Mantas Luko s evi c ius, Dan Popovici, and Udo Siewert. Optimization and applications of Echo State Networks with leaky-integrator neurons. Neural Networks, 20 0 (3): 0 335--352, 2007

  11. [22]

    No free prune: Information-theoretic barriers to pruning at initialization

    Tanishq Kumar, Kevin Luo, and Mark Sellke. No free prune: Information-theoretic barriers to pruning at initialization. In Proceedings of the 41st International Conference on Machine Learning, volume 235 of Proceedings of Machine Learning Research, pages 25662--25681. PMLR, 202...

  12. [23]

    ZipLM : Inference-aware structured pruning of language models

    Eldar Kurti \'c , Elias Frantar, and Dan Alistarh. ZipLM : Inference-aware structured pruning of language models. In Advances in Neural Information Processing Systems, volume 36, pages 65597--65617. Curran Associates, Inc., 2023. URL https://proceedings.neurips.cc/paper_files/...

  13. [24]

    Efficient behavior of small-world networks

    Vito Latora and Massimo Marchiori. Efficient behavior of small-world networks. Physical Review Letters, 87 0 (19): 0 198701, 2001

  14. [26]

    Broad Echo State Network with reservoir pruning for nonstationary time series prediction

    Wenjie Liu, Yuting Bai, Xuebo Jin, Xiaoyi Wang, Tingli Su, and Jianlei Kong. Broad Echo State Network with reservoir pruning for nonstationary time series prediction. Computational Intelligence and Neuroscience, 2022 0 (1): 0 3672905, 2022

  15. [27]

    Reservoir computing approaches to recurrent neural network training

    Mantas Luko s evi c ius and Herbert Jaeger. Reservoir computing approaches to recurrent neural network training. Computer Science Review, 3 0 (3): 0 127--149, 2009

  16. [28]

    Convolutional multitimescale Echo State Network

    Qianli Ma, Enhuan Chen, Zhenxi Lin, Jiangyue Yan, Zhiwen Yu, and Wing WY Ng. Convolutional multitimescale Echo State Network . IEEE Transactions on Cybernetics , 51 0 (3): 0 1613--1625, 2019

  17. [29]

    Oscillation and chaos in physiological control systems

    Michael C Mackey and Leon Glass. Oscillation and chaos in physiological control systems. Science, 197 0 (4300): 0 287--289, 1977

  18. [30]

    Echo State Property linked to an input: Exploring a fundamental characteristic of recurrent neural networks

    Gandhi Manjunath and Herbert Jaeger. Echo State Property linked to an input: Exploring a fundamental characteristic of recurrent neural networks. Neural Computation, 25 0 (3): 0 671--696, 2013

  19. [32]

    NREL wind integration national dataset toolkit

    National Renewable Energy Laboratory . NREL wind integration national dataset toolkit. https://www.nrel.gov/grid/wind-toolkit.html, 2024 a . Accessed: 2026-03-31

  20. [33]

    National solar radiation database ( NSRDB )

    National Renewable Energy Laboratory . National solar radiation database ( NSRDB ). https://nsrdb.nrel.gov/, 2024 b . Accessed: 2026-03-31

  21. [34]

    A high-performance deep reservoir computer experimentally demonstrated with ion-gating reservoirs

    Daiki Nishioka, Takashi Tsuchiya, Masataka Imura, Yasuo Koide, Tohru Higuchi, and Kazuya Terabe. A high-performance deep reservoir computer experimentally demonstrated with ion-gating reservoirs. Communications Engineering, 3 0 (1): 0 81, 2024

  22. [35]

    Fantastic weights and how to find them: Where to prune in dynamic sparse training

    Aleksandra Nowak, Bram Grooten, Decebal Constantin Mocanu, and Jacek Tabor. Fantastic weights and how to find them: Where to prune in dynamic sparse training. In Advances in Neural Information Processing Systems, volume 36, pages 55160--55192. Curran Associates, Inc., 2023. UR...

  23. [36]

    Minimum complexity Echo State Network

    Ali Rodan and Peter Tino. Minimum complexity Echo State Network . IEEE Transactions on Neural Networks , 22 0 (1): 0 131--144, 2010

  24. [37]

    An effective criterion for pruning reservoir's connections in Echo State Networks

    Simone Scardapane, Gabriele Nocco, Danilo Comminiello, Michele Scarpiniti, and Aurelio Uncini. An effective criterion for pruning reservoir's connections in Echo State Networks . In 2014 International Joint Conference on Neural Networks ( IJCNN ) , pages 1205--1212. IEEE, 2014

  25. [38]

    Echo State Network optimization: A systematic literature review

    Rebh Soltani, Emna Benmohamed, and Hela Ltifi. Echo State Network optimization: A systematic literature review. Neural Processing Letters, 55 0 (8): 0 10251--10285, 2023

  26. [39]

    Deep Echo State Network pruning algorithm based on detrended multiple cross-correlation

    Xiaochuan Sun, Yu Wang, et al. Deep Echo State Network pruning algorithm based on detrended multiple cross-correlation. Journal of Zhengzhou University (Engineering Science), 45 0 (4): 0 38--45, 2024

  27. [40]

    An experimental unification of reservoir computing methods

    David Verstraeten, Benjamin Schrauwen, Michiel d'Haene, and Dirk Stroobandt. An experimental unification of reservoir computing methods. Neural Networks, 20 0 (3): 0 391--403, 2007

  28. [42]

    Chain-structure Echo State Network with stochastic optimization: Methodology and application

    Zhou Wu, Qian Li, and Haijun Zhang. Chain-structure Echo State Network with stochastic optimization: Methodology and application. IEEE Transactions on Neural Networks and Learning Systems , 33 0 (5): 0 1974--1985, 2021

  29. [44]

    Topology-aware network pruning using multi-stage graph embedding and reinforcement learning

    Sixing Yu, Arya Mazaheri, and Ali Jannesari. Topology-aware network pruning using multi-stage graph embedding and reinforcement learning. In Proceedings of the 39th International Conference on Machine Learning, volume 162 of Proceedings of Machine Learning Research, pages 2565...

  30. [45]

    One-shot pruning of recurrent neural networks by Jacobian spectrum evaluation

    Matthew Shunshi Zhang and Bradly Stadie. One-shot pruning of recurrent neural networks by Jacobian spectrum evaluation. arXiv e-prints, page arXiv:1912.00120, 2019

  31. [46]

    Information dynamics in neuromorphic nanowire networks

    Ruomin Zhu, Joel Hochstetter, Alon Loeffler, Adrian Diaz-Alvarez, Tomonobu Nakayama, Joseph T Lizier, and Zdenka Kuncic. Information dynamics in neuromorphic nanowire networks. Scientific Reports, 11 0 (1): 0 13047, 2021

  32. [47]

    Scholarpedia , volume =

    Jaeger, Herbert , title =. Scholarpedia , volume =

  33. [48]

    Jaeger, Herbert , title =

  34. [49]

    Reservoir Computing Approaches to Recurrent Neural Network Training , journal =

    Luko. Reservoir Computing Approaches to Recurrent Neural Network Training , journal =. 2009 , publisher =

  35. [50]

    Neural Networks , volume =

    Verstraeten, David and Schrauwen, Benjamin and d'Haene, Michiel and Stroobandt, Dirk , title =. Neural Networks , volume =. 2007 , publisher =

  36. [51]

    2019 , publisher =

    Ma, Qianli and Chen, Enhuan and Lin, Zhenxi and Yan, Jiangyue and Yu, Zhiwen and Ng, Wing WY , title =. 2019 , publisher =

  37. [52]

    Real-Time Computing without Stable States: A New Framework for Neural Computation Based on Perturbations , journal =

    Maass, Wolfgang and Natschl. Real-Time Computing without Stable States: A New Framework for Neural Computation Based on Perturbations , journal =. 2002 , publisher =

  38. [53]

    Optimization and Applications of

    Jaeger, Herbert and Luko. Optimization and Applications of. Neural Networks , volume =. 2007 , publisher =

  39. [54]

    Training

    L. Training. Cognitive Computation , volume =. 2017 , publisher =

  40. [55]

    2014 International Joint Conference on Neural Networks (

    Scardapane, Simone and Nocco, Gabriele and Comminiello, Danilo and Scarpiniti, Michele and Uncini, Aurelio , title =. 2014 International Joint Conference on Neural Networks (. 2014 , organization =

  41. [56]

    2012 , institution =

    Jaeger, Herbert , title =. 2012 , institution =

  42. [57]

    arXiv preprint arXiv:1412.3555 , year =

    Chung, Junyoung and Gulcehre, Caglar and Cho, KyungHyun and Bengio, Yoshua , title =. arXiv preprint arXiv:1412.3555 , year =

  43. [58]

    Frontiers in Computational Neuroscience , volume =

    Schubert, Fabian and Gros, Claudius , title =. Frontiers in Computational Neuroscience , volume =. 2021 , publisher =

  44. [59]

    Neural Networks , volume =

    Yildiz, Izzet B and Jaeger, Herbert and Kiebel, Stefan J , title =. Neural Networks , volume =. 2012 , publisher =

  45. [60]

    Neural Computation , volume =

    Ozturk, Mustafa C and Xu, Dongming and Principe, Jose C , title =. Neural Computation , volume =. 2007 , publisher =

  46. [61]

    2010 , publisher =

    Rodan, Ali and Tino, Peter , title =. 2010 , publisher =

  47. [62]

    Boedecker, Joschka and Obst, Oliver and Mayer, Norbert Michael and Asada, Minoru , title =

  48. [63]

    Engineering Applications of Artificial Intelligence , volume =

    Wang, Zhigang and Zeng, Yu-Rong and Wang, Sirui and Wang, Lin , title =. Engineering Applications of Artificial Intelligence , volume =. 2019 , publisher =

  49. [64]

    Proceedings of the

    Zhao, Ruizhe and Luk, Wayne , title =. Proceedings of the

  50. [65]

    Journal of Intelligent & Fuzzy Systems , volume =

    Sun, Xiaochuan and Wang, Yu and Hao, Mingxiang and Li, Yingqi and Huang, Tianyu , title =. Journal of Intelligent & Fuzzy Systems , volume =. 2024 , publisher =

  51. [66]

    Proceedings of the VLDB Endowment , volume =

    Vatter, Jana and Rochau, Maurice L and Mayer, Ruben and Jacobsen, Hans-Arno , title =. Proceedings of the VLDB Endowment , volume =

  52. [67]

    Computational Intelligence and Neuroscience , volume =

    Liu, Wenjie and Bai, Yuting and Jin, Xuebo and Wang, Xiaoyi and Su, Tingli and Kong, Jianlei , title =. Computational Intelligence and Neuroscience , volume =. 2022 , publisher =

  53. [68]

    2023 , publisher =

    He, Yang and Xiao, Lingao , title =. 2023 , publisher =

  54. [69]

    Journal of Machine Learning Research , volume =

    Hoefler, Torsten and Alistarh, Dan and Ben-Nun, Tal and Dryden, Nikoli and Peste, Alexandra , title =. Journal of Machine Learning Research , volume =

  55. [70]

    Cognitive Computation , volume =

    Tmamna, Jihene and Ayed, Emna Ben and Fourati, Rahma and Gogate, Mandar and Arslan, Tughrul and Hussain, Amir and Ayed, Mounir Ben , title =. Cognitive Computation , volume =. 2024 , publisher =

  56. [71]

    Measurement Science and Technology , volume =

    Huang, Jian and Wang, Fan and Yang, Xu and Li, Qing , title =. Measurement Science and Technology , volume =. 2023 , publisher =

  57. [72]

    Social Choice and Welfare , volume =

    Bloch, Francis and Jackson, Matthew O and Tebaldi, Pietro , title =. Social Choice and Welfare , volume =. 2023 , publisher =

  58. [73]

    PLOS ONE , volume =

    Joyce, Karen E and Laurienti, Paul J and Burdette, Jonathan H and Hayasaka, Satoru , title =. PLOS ONE , volume =. 2010 , publisher =

  59. [74]

    Neural Processing Letters , volume =

    Soltani, Rebh and Benmohamed, Emna and Ltifi, Hela , title =. Neural Processing Letters , volume =. 2023 , publisher =

  60. [75]

    Social Networks , volume =

    Borgatti, Stephen P , title =. Social Networks , volume =. 2005 , publisher =

  61. [76]

    Scientific Reports , volume =

    Zhu, Ruomin and Hochstetter, Joel and Loeffler, Alon and Diaz-Alvarez, Adrian and Nakayama, Tomonobu and Lizier, Joseph T and Kuncic, Zdenka , title =. Scientific Reports , volume =. 2021 , publisher =

  62. [77]

    PLOS ONE , volume =

    Cavallaro, Lucia and De Meo, Pasquale and Fiumara, Giacomo and Liotta, Antonio , title =. PLOS ONE , volume =. 2024 , publisher =

  63. [78]

    2018 , publisher =

    Newman, Mark , title =. 2018 , publisher =

  64. [79]

    Science , volume =

    Mackey, Michael C and Glass, Leon , title =. Science , volume =. 1977 , publisher =

  65. [80]

    Advances in Chemical Physics , volume =

    Bartsch, Thomas and Moix, Jeremy M and Hernandez, Rigoberto and Kawai, Shinnosuke and Uzer, Turgay , title =. Advances in Chemical Physics , volume =

  66. [81]

    Matzner, Filip and Mráz, František , title =

  67. [82]

    Proceedings of the 2019 11th International Conference on Machine Learning and Computing , year =

    Liu, Chongdang and Yao, Rong and Zhang, Linxuan and Liao, Yuan , title =. Proceedings of the 2019 11th International Conference on Machine Learning and Computing , year =

  68. [83]

    arXiv preprint arXiv:2403.19806 , year =

    Goswami, Debdipta , title =. arXiv preprint arXiv:2403.19806 , year =

  69. [84]

    Proceedings of the 33rd International Conference on Machine Learning , pages =

    Couillet, Romain and Debbah, Merouane and Silverstein, Jack W , title =. Proceedings of the 33rd International Conference on Machine Learning , pages =. 2016 , publisher =

  70. [85]

    Social Networks , volume =

    Freeman, Linton C , title =. Social Networks , volume =. 1979 , publisher =

  71. [86]

    The Science of Networks , journal =

    Barab. The Science of Networks , journal =

  72. [87]

    Physical Review Letters , volume =

    Latora, Vito and Marchiori, Massimo , title =. Physical Review Letters , volume =. 2001 , publisher =

  73. [88]

    Social Networks , volume =

    Opsahl, Tore and Agneessens, Filip and Skvoretz, John , title =. Social Networks , volume =. 2010 , publisher =

  74. [89]

    Journal of Mathematical Sociology , volume =

    Brandes, Ulrik , title =. Journal of Mathematical Sociology , volume =. 2001 , publisher =

  75. [90]

    Psychometrika , volume =

    Sabidussi, Gert , title =. Psychometrika , volume =. 1966 , publisher =

  76. [91]

    Physica A: Statistical Mechanics and its Applications , volume =

    Zhao, Zhi and Sun, Weigang , title =. Physica A: Statistical Mechanics and its Applications , volume =. 2014 , publisher =

  77. [92]

    and Rout, A

    Rout, D. and Rout, A. , title =. International Journal of Applied Research , volume =

  78. [93]

    and Stoffer, David S

    Shumway, Robert H. and Stoffer, David S. , title =

  79. [94]

    and Nobre, C

    Andrade de Oliveira, F. and Nobre, C. , title =. Neural Computing and Applications , volume =

  80. [95]

    , title =

    Jaeger, H. , title =

  81. [96]

    Reservoir Computing Approaches to Recurrent Neural Network Training , journal =

    Luko. Reservoir Computing Approaches to Recurrent Neural Network Training , journal =

  82. [97]

    and Zhang, H

    Jingpei, L. and Zhang, H. , title =. Neural Networks , volume =

  83. [98]

    and Min, X

    Shu, W. and Min, X. , title =

  84. [99]

    Newman, Mark E. J. , title =

  85. [100]

    Fletcher, J. M. and Wennekers, T. , title =. Neuroinformatics , volume =

  86. [101]

    Nachrichten von der Gesellschaft der Wissenschaften zu G

    Landau, Edmund , title =. Nachrichten von der Gesellschaft der Wissenschaften zu G. 1895 , note =

  87. [102]

    Eigenvector Centrality , year =

  88. [103]

    , title =

    Freeman, Linton C. , title =. Sociometry , volume =. 1977 , doi =

  89. [104]

    Psychometrika , volume =

    Sabidussi, Gert , title =. Psychometrika , volume =. 1966 , doi =

  90. [105]

    American Journal of Sociology , volume =

    Bonacich, Phillip , title =. American Journal of Sociology , volume =. 1987 , doi =

  91. [106]

    Newman, Mark E. J. , title =. 2010 , isbn =

  92. [107]

    Neurocomputing , volume =

    Dutoit, Xavier and Schrauwen, Benjamin and others , title =. Neurocomputing , volume =. 2009 , publisher =

  93. [108]

    Li, Chen and Zhang, Hao and others , title =

  94. [109]

    2021 , publisher =

    Wu, Zhou and Li, Qian and Zhang, Haijun , title =. 2021 , publisher =

  95. [110]

    Journal of Zhengzhou University (Engineering Science) , volume =

    Sun, Xiaochuan and Wang, Yu and others , title =. Journal of Zhengzhou University (Engineering Science) , volume =

  96. [111]

    Communications Engineering , volume =

    Nishioka, Daiki and Tsuchiya, Takashi and Imura, Masataka and Koide, Yasuo and Higuchi, Tohru and Terabe, Kazuya , title =. Communications Engineering , volume =. 2024 , publisher =

  97. [112]

    Universality in Chaos , edition =

    Lorenz, Edward N , title =. Universality in Chaos , edition =. 2017 , publisher =

  98. [113]

    2024 , howpublished =

    Global Surface Summary of the Day (. 2024 , howpublished =

  99. [114]

    2024 , howpublished =

  100. [115]

    2024 , howpublished =

    National Solar Radiation Database (. 2024 , howpublished =

  101. [116]

    Residential Electricity Consumption Data , year =

  102. [117]

    2006 , howpublished =

    Hebrail, Georges and Berard, Alice , title =. 2006 , howpublished =

  103. [118]

    Wind Turbine

    Berker I. Wind Turbine. 2020 , howpublished =

  104. [119]

    Climate Data Online , year =

  105. [120]

    arXiv preprint arXiv:2603.20684 , year =

    Laudari, Sudip , title =. arXiv preprint arXiv:2603.20684 , year =

  106. [121]

    Cognitive Computation , volume =

    Gallicchio, Claudio and Micheli, Alessio , title =. Cognitive Computation , volume =. 2017 , publisher =

  107. [122]

    arXiv e-prints , pages =

    Zhang, Matthew Shunshi and Stadie, Bradly , title =. arXiv e-prints , pages =

  108. [123]

    2016 , publisher =

    Bianchi, Filippo Maria and Livi, Lorenzo and Alippi, Cesare , title =. 2016 , publisher =

  109. [124]

    Neural Networks , volume =

    Chatzikonstantinou, Christos and Konstantinidis, Dimitrios and Dimitropoulos, Kosmas and Daras, Petros , title =. Neural Networks , volume =. 2021 , publisher =

  110. [125]

    arXiv preprint arXiv:1811.10892 , year =

    Gallicchio, Claudio , title =. arXiv preprint arXiv:1811.10892 , year =

  111. [126]

    Physica D: Nonlinear Phenomena , volume =

    Ceni, Andrea and Ashwin, Peter and Livi, Lorenzo and Postlethwaite, Claire , title =. Physica D: Nonlinear Phenomena , volume =. 2020 , publisher =

  112. [127]

    Neurocomputing , volume =

    Gallicchio, Claudio and Micheli, Alessio and Silvestri, Luca , title =. Neurocomputing , volume =. 2018 , publisher =

  113. [128]

    Neural Computation , volume =

    Manjunath, Gandhi and Jaeger, Herbert , title =. Neural Computation , volume =. 2013 , publisher =

  114. [129]

    , title =

    Han, Song and Pool, Jeff and Tran, John and Dally, William J. , title =. Advances in Neural Information Processing Systems , volume =. 2015 , publisher =

  115. [130]

    International Conference on Learning Representations , year =

    Frankle, Jonathan and Carbin, Michael , title =. International Conference on Learning Representations , year =

  116. [131]

    Proceedings of Machine Learning and Systems , volume =

    Blalock, Davis and Gonzalez Ortiz, Jose Javier and Frankle, Jonathan and Guttag, John , title =. Proceedings of Machine Learning and Systems , volume =. 2020 , publisher =

  117. [132]

    Advances in Neural Information Processing Systems , volume =

    Nowak, Aleksandra and Grooten, Bram and Mocanu, Decebal Constantin and Tabor, Jacek , title =. Advances in Neural Information Processing Systems , volume =. 2023 , publisher =

  118. [133]

    Proceedings of the 41st International Conference on Machine Learning , volume =

    Kumar, Tanishq and Luo, Kevin and Sellke, Mark , title =. Proceedings of the 41st International Conference on Machine Learning , volume =. 2024 , publisher =

  119. [134]

    Proceedings of the 39th International Conference on Machine Learning , volume =

    Yu, Sixing and Mazaheri, Arya and Jannesari, Ali , title =. Proceedings of the 39th International Conference on Machine Learning , volume =. 2022 , publisher =

  120. [135]

    Advances in Neural Information Processing Systems , volume =

    El Halabi, Marwa and Srinivas, Suraj and Lacoste-Julien, Simon , title =. Advances in Neural Information Processing Systems , volume =. 2022 , publisher =

  121. [136]

    Advances in Neural Information Processing Systems , volume =

    Wright, Dustin and Igel, Christian and Selvan, Raghavendra , title =. Advances in Neural Information Processing Systems , volume =. 2024 , publisher =. doi:10.52202/079017-2045 , url =

  122. [137]

    Proceedings of the 39th International Conference on Machine Learning , volume =

    Frantar, Elias and Alistarh, Dan , title =. Proceedings of the 39th International Conference on Machine Learning , volume =. 2022 , publisher =

  123. [138]

    Advances in Neural Information Processing Systems , volume =

    Kurti. Advances in Neural Information Processing Systems , volume =. 2023 , publisher =

  124. [139]

    Proceedings of the 39th International Conference on Machine Learning , volume =

    He, Zheng and Xie, Zeke and Zhu, Quanzhi and Qin, Zengchang , title =. Proceedings of the 39th International Conference on Machine Learning , volume =. 2022 , publisher =

  125. [140]

    Chaos: An Interdisciplinary Journal of Nonlinear Science , volume =

    Yadav, Manish and Stender, Merten , title =. Chaos: An Interdisciplinary Journal of Nonlinear Science , volume =. 2025 , publisher =. doi:10.1063/5.0273535 , url =

  126. [141]

    Chaos: An Interdisciplinary Journal of Nonlinear Science , volume =

    Martinuzzi, Francesco , title =. Chaos: An Interdisciplinary Journal of Nonlinear Science , volume =. 2025 , publisher =. doi:10.1063/5.0288751 , url =

  127. [142]

    Advances in Neural Information Processing Systems , volume =

    Gwak, Minseon and Moon, Seongrok and Ko, Joohwan and Park, PooGyeon , title =. Advances in Neural Information Processing Systems , volume =. 2024 , publisher =. doi:10.52202/079017-0339 , url =

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.