pith. sign in

arxiv: 2509.11612 · v3 · submitted 2025-09-15 · 💻 cs.LG

Topology Structure Optimization of Reservoirs Using GLMY Homology

Yu Chen , Shengwei Wang , Hongwei Lin This is my paper

Pith reviewed 2026-05-18 15:55 UTC · model grok-4.3

classification 💻 cs.LG
keywords reservoir computingGLMY homologytopology optimizationpersistent homologytime series processingnetwork structureecho state networks
0
0 comments X

The pith

Reservoir performance for time series improves when minimal cycles in one-dimensional GLMY homology groups are modified.

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

This paper uses persistent GLMY homology to analyze the topology of reservoir networks and shows that their performance on time series tasks is closely tied to the one-dimensional homology groups. The authors create an optimization method that targets and changes the minimal representative cycles of these groups to adjust the network structure. Experiments demonstrate that the resulting performance gains depend on both the modified topology and the periodicity present in the training data. A reader would care because this supplies a concrete mathematical handle on reservoir design that goes beyond trial-and-error adjustments of connectivity.

Core claim

The reservoir performance is closely related to the one-dimensional GLMY homology groups. A reservoir structure optimization method is developed by modifying the minimal representative cycles of one-dimensional GLMY homology groups. Experiments validate that the performance of reservoirs is jointly influenced by the reservoir structure and the periodicity of the dataset.

What carries the argument

One-dimensional GLMY homology groups and their minimal representative cycles, which encode topological features of the reservoir and serve as the direct targets for structural modification.

If this is right

  • Modifying minimal representative cycles of one-dimensional GLMY homology groups produces measurable gains in reservoir performance.
  • Reservoir performance is jointly determined by network structure and the periodicity of the input dataset.
  • GLMY homology supplies a systematic way to diagnose and adjust reservoir topologies.

Where Pith is reading between the lines

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

  • The same cycle-modification approach could be tested on other recurrent architectures that share reservoir-like dynamics.
  • Higher-dimensional GLMY homology groups might expose additional structural levers once the one-dimensional case is established.
  • Datasets with controlled periodic components could be used to isolate how much of the performance change is attributable to the homology adjustment alone.

Load-bearing premise

Changing the minimal representative cycles of the one-dimensional GLMY homology groups produces a causal gain in computational capability rather than incidental side effects on other network properties.

What would settle it

An experiment that modifies only the minimal representative cycles of the one-dimensional GLMY homology groups and then measures no improvement or a drop in time-series prediction accuracy while holding connectivity density and other parameters fixed.

Figures

Figures reproduced from arXiv: 2509.11612 by Hongwei Lin, Shengwei Wang, Yu Chen.

Figure 1
Figure 1. Figure 1: Illustration of the flowchart of our method and experimental procedures. for forecasting stock market prices, where comparative experiments against several classical models—including long short-term memory (LSTM) networks—showed the RC model consistently achieving lower mean-squared error (MSE) and mean absolute error (MAE), alongside higher 𝑅2 scores. This work also provided insights into how dis￾tinct re… view at source ↗
Figure 2
Figure 2. Figure 2: (a) Example of an one-dimensional persistence diagram. (b) An illustration of the confidence set (red area) of (a) [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Structure of Reservoir Computing Model computing model is illustrated in [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Illustration of compatibly modifying cycles into rings. (a) A digraph has two rings on the left and right. The middle cycle (red) is not a ring. (b) By changing the direction of 𝐴𝐷, the middle cycle becomes a ring 𝐴𝐵𝐶𝐷 compatible with the other rings. (c) The case where the middle cycle cannot be changed into a ring without destroying the other two rings, since either 𝐴𝐵 or 𝐷𝐶 would need to have its direct… view at source ↗
Figure 5
Figure 5. Figure 5: Illustration of four datasets, and their Time-delay embedding results in ℝ3 and the one-dimensional persistence diagrams. (a) Result of Mackey-Glass Data. (b) Result of MSO Data. (c) Result of Lorenz System Data. (d) Result of NARMA Data. Algorithm 1 Reservoir Optimization by GLMY Homology Input: A digraph 𝐺 = (𝑉 , 𝐸) representing a reservoir. 1: Computing the persistent GLMY homology of 𝐺 and get the 1-PD… view at source ↗
Figure 6
Figure 6. Figure 6: Average memory capacity (MC) tested on four datasets before and after optimization. Here we compared three initialization approaches. 5.3. Memory capacity analysis In the subsequent experiments, the reservoir digraph’s node number is still set as 500. For the Random ini￾tial method, the reservoir sparsity is set to 0.99 (i.e., 1% connectivity), which is a typical setting in practice [16]. This configuratio… view at source ↗
Figure 7
Figure 7. Figure 7: Illustration of the prediction results on four datasets. (a)–(d) are results of the Mackey-Glass, MSO, Lorenz system, and NARMA dataset, respectively. Figures on the top are prediction results of the initial randomly generated reservoir, with black being the real data and red being the predicted values. It can be seen that after a period of time, the reservoir loses its predictive ability; Figures on the b… view at source ↗
Figure 8
Figure 8. Figure 8: Average RMSE tested on four datasets before and after optimizing the reservoir digraph. Here we compared three initialization approaches. edges (which means their corresponding fundamental cycle submatrices will share common columns), the increased number of rings is not proportional to improved orthogonal￾ity. It is still worthy studying more efficient and reasonable method for improving orthogonality. Ad… view at source ↗
read the original abstract

Reservoir is an efficient network for time series processing. It is well known that network structure is one of the determinants of its performance. However, the topology structure of reservoirs, as well as their performance, is hard to analyzed, due to the lack of suitable mathematical tools. In this paper, we study the topology structure of reservoirs using persistent GLMY homology theory, and develop a method to improve its performance. Specifically, it is found that the reservoir performance is closely related to the one-dimensional GLMY homology groups. Then, we develop a reservoir structure optimization method by modifying the minimal representative cycles of one-dimensional GLMY homology groups. Finally, by experiments, it is validated that the performance of reservoirs is jointly influenced by the reservoir structure and the periodicity of the dataset.

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 paper applies persistent GLMY homology to reservoir computing networks for time series processing. It claims that performance is closely related to the one-dimensional GLMY homology groups, introduces a structure optimization method that modifies the minimal representative cycles of these groups, and reports experiments showing that reservoir performance is jointly influenced by structure and dataset periodicity.

Significance. If the claimed causal relationship between targeted GLMY cycle modifications and performance gains can be isolated from generic structural changes, the work would introduce a new topological tool for reservoir design that leverages algebraic invariants of directed graphs, potentially advancing structure-aware reservoir computing beyond heuristic or spectral methods.

major comments (2)
  1. [Experiments] The experimental validation does not include ablation controls that apply random or degree-preserving perturbations of comparable magnitude while leaving the one-dimensional GLMY homology unchanged; without such controls it is impossible to attribute observed gains specifically to the homology modification rather than incidental changes in connectivity, spectral radius, or memory capacity.
  2. [Experiments] The manuscript provides no description of the datasets, baseline reservoir constructions, hyperparameter settings, number of trials, or statistical tests used to support the performance claims and the joint-influence statement with dataset periodicity.
minor comments (2)
  1. [Method] Notation for GLMY homology groups and minimal representative cycles should be defined more explicitly with reference to the underlying chain complexes and boundary maps.
  2. [Experiments] Figure captions and axis labels in the experimental results should include error bars or confidence intervals to convey variability across runs.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which highlight important aspects of experimental rigor. We address each major comment below and commit to revisions that strengthen the attribution of performance gains to the GLMY homology modifications while providing full experimental details.

read point-by-point responses

  1. Referee: [Experiments] The experimental validation does not include ablation controls that apply random or degree-preserving perturbations of comparable magnitude while leaving the one-dimensional GLMY homology unchanged; without such controls it is impossible to attribute observed gains specifically to the homology modification rather than incidental changes in connectivity, spectral radius, or memory capacity.

    Authors: We agree that isolating the causal effect of the targeted GLMY cycle modifications requires controls that preserve the one-dimensional homology groups. In the revised manuscript we will add ablation experiments applying random and degree-preserving perturbations of comparable magnitude that leave the one-dimensional GLMY homology unchanged. Performance differences between these controls and our homology-guided modifications will be reported, together with measurements of spectral radius and memory capacity, to demonstrate that gains arise specifically from the homology modifications rather than generic connectivity changes. revision: yes

  2. Referee: [Experiments] The manuscript provides no description of the datasets, baseline reservoir constructions, hyperparameter settings, number of trials, or statistical tests used to support the performance claims and the joint-influence statement with dataset periodicity.

    Authors: We acknowledge that the original manuscript omitted sufficient experimental details. The revised version will include a dedicated Experimental Setup section describing the datasets (including their periodicity properties), baseline reservoir constructions (random, small-world, and scale-free networks), hyperparameter values (reservoir size, spectral radius, leaking rate, input scaling), number of independent trials (20 runs with distinct random seeds), and statistical tests (mean and standard deviation with paired t-tests for significance). These additions will substantiate the reported joint influence of structure and dataset periodicity. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is empirical and self-contained

full rationale

The paper's chain begins with an empirical observation that reservoir performance relates to one-dimensional GLMY homology groups, proceeds to a proposed optimization method that modifies minimal representative cycles, and concludes with experimental validation of joint influence by structure and dataset periodicity. None of these steps reduce by definition or construction to prior inputs; the homology-performance link is presented as a discovered relation rather than a definitional identity, the modification is an explicit algorithmic choice, and validation rests on external experiments rather than fitted parameters renamed as predictions or self-citation chains. The derivation therefore remains independent of its own outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the unverified premise that one-dimensional GLMY homology groups encode performance-relevant structure; this premise is treated as a discovery but functions as the load-bearing domain assumption for the entire optimization method.

axioms (1)
  • domain assumption Reservoir performance is closely related to the one-dimensional GLMY homology groups.
    This stated relation is used to justify modifying minimal representative cycles as an optimization strategy.

pith-pipeline@v0.9.0 · 5654 in / 1142 out tokens · 57494 ms · 2026-05-18T15:55:25.095090+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

46 extracted references · 46 canonical work pages · 2 internal anchors

  1. [1]

    LeCun, Y

    Y . LeCun, Y . Bengio, G. Hinton, Deep learning, nature 521 (7553) (2015) 436–444

    work page 2015

  2. [2]

    Goodfellow, Y

    I. Goodfellow, Y . Bengio, A. Courville, Y . Bengio, Deep learning, V ol. 1, MIT press Cambridge, 2016

    work page 2016

  3. [3]

    Hornik, M

    K. Hornik, M. Stinchcombe, H. White, Multilayer feedforward net- works are universal approximators, Neural networks 2 (5) (1989) 359–366

    work page 1989

  4. [4]

    Gaier, D

    A. Gaier, D. Ha, Weight agnostic neural networks, Advances in neural information processing systems 32 (2019). :Preprint submitted to Elsevier Page 15 of 16

    work page 2019

  5. [5]

    Montavon, W

    G. Montavon, W. Samek, K.-R. Müller, Methods for interpreting and understanding deep neural networks, Digital signal processing 73 (2018) 1–15

    work page 2018

  6. [6]

    Zhang, Z

    B. Zhang, Z. He, H. Lin, A comprehensive review of deep neural net- work interpretation using topological data analysis, Neurocomputing 609 (2024) 128513

    work page 2024

  7. [7]

    Lukoševi ˇcius, H

    M. Lukoševi ˇcius, H. Jaeger, Reservoir computing approaches to recurrent neural network training, Computer science review 3 (3) (2009) 127–149

    work page 2009

  8. [8]

    echo state

    H. Jaeger, The “echo state” approach to analysing and training recurrent neural networks-with an erratum note, Bonn, Germany: German national research center for information technology gmd technical report 148 (34) (2001) 13

    work page 2001

  9. [9]

    Pathak, B

    J. Pathak, B. Hunt, M. Girvan, Z. Lu, E. Ott, Model-free prediction of large spatiotemporally chaotic systems from data: A reservoir computing approach, Physical review letters 120 (2) (2018) 024102

    work page 2018

  10. [10]

    Homologies of path complexes and digraphs

    A. Grigor’yan, Y . Lin, Y . Muranov, S.-T. Yau, Homologies of path complexes and digraphs, arXiv preprint arXiv:1207.2834 (2012)

    work page internal anchor Pith review Pith/arXiv arXiv 2012

  11. [11]

    Chowdhury, F

    S. Chowdhury, F. Mémoli, Persistent path homology of directed networks, in: Proceedings of the Twenty-Ninth Annual ACM-SIAM Symposium on Discrete Algorithms, SIAM, 2018, pp. 1152–1169

    work page 2018

  12. [12]

    Gallicchio, A

    C. Gallicchio, A. Micheli, Reservoir topology in deep echo state networks, in: Artificial Neural Networks and Machine Learning– ICANN 2019: Workshop and Special Sessions: 28th International Conference on Artificial Neural Networks, Munich, Germany, September 17–19, 2019, Proceedings 28, Springer, 2019, pp. 62–75

    work page 2019

  13. [13]

    Stockdill, K

    A. Stockdill, K. Neshatian, Restricted echo state networks, in: AI 2016: Advances in Artificial Intelligence: 29th Australasian Joint Conference, Hobart, TAS, Australia, December 5-8, 2016, Proceed- ings 29, Springer, 2016, pp. 555–560

    work page 2016

  14. [14]

    X. Yao, Z. Wang, H. Zhang, Prediction and identification of discrete- time dynamic nonlinear systems based on adaptive echo state net- work, Neural Networks 113 (2019) 11–19

    work page 2019

  15. [15]

    M. D. Skowronski, J. G. Harris, Noise-robust automatic speech recognition using a predictive echo state network, IEEE Transactions on Audio, Speech, and Language Processing 15 (5) (2007) 1724– 1730

    work page 2007

  16. [16]

    Jaeger, H

    H. Jaeger, H. Haas, Harnessing nonlinearity: Predicting chaotic systems and saving energy in wireless communication, science 304 (5667) (2004) 78–80

    work page 2004

  17. [17]

    M. R. E. U. Shougat, X. Li, T. Mollik, E. Perkins, A hopf physical reservoir computer, Scientific Reports 11 (1) (2021) 19465

    work page 2021

  18. [18]

    W.-J. Wang, Y . Tang, J. Xiong, Y .-C. Zhang, Stock market index prediction based on reservoir computing models, Expert Systems with Applications 178 (2021) 115022

    work page 2021

  19. [19]

    H. Cui, X. Liu, L. Li, The architecture of dynamic reservoir in the echo state network, Chaos: An Interdisciplinary Journal of Nonlinear Science 22 (3) (2012)

    work page 2012

  20. [20]

    Roeschies, C

    B. Roeschies, C. Igel, Structure optimization of reservoir networks, Logic Journal of IGPL 18 (5) (2010) 635–669

    work page 2010

  21. [21]

    Sun, H.-y

    X.-c. Sun, H.-y. Cui, R.-p. Liu, J.-y. Chen, Y .-j. Liu, Modeling deterministic echo state network with loop reservoir, Journal of Zhejiang University SCIENCE C 13 (9) (2012) 689–701

    work page 2012

  22. [22]

    Grigor’yan, R

    A. Grigor’yan, R. Jimenez, Y . Muranov, S.-T. Yau, Homology of path complexes and hypergraphs, Topology and its Applications 267 (2019) 106877

    work page 2019

  23. [23]

    Homotopy theory for digraphs

    A. Grigor’yan, Y . Lin, Y . Muranov, S.-T. Yau, Homotopy theory for digraphs, arXiv preprint arXiv:1407.0234 (2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  24. [24]

    Y . Lin, C. Wang, S.-T. Yau, Discrete morse theory on digraphs, arXiv preprint arXiv:2102.10518 (2021)

    work page arXiv 2021

  25. [25]

    Grigor’yan, Y

    A. Grigor’yan, Y . Muranov, V . Vershinin, S.-T. Yau, Path homology theory of multigraphs and quivers, in: Forum mathematicum, V ol. 30, De Gruyter, 2018, pp. 1319–1337

    work page 2018

  26. [26]

    T. K. Dey, T. Li, Y . Wang, An efficient algorithm for 1-dimensional (persistent) path homology, Discrete & Computational Geometry 68 (4) (2022) 1102–1132

    work page 2022

  27. [27]

    L. Feng, H. Gong, S. Zhang, X. Liu, Y . Wang, J. Che, A. Dong, C. H. Griffin, C. Gragnoli, J. Wu, et al., Hypernetwork modeling and topol- ogy of high-order interactions for complex systems, Proceedings of the National Academy of Sciences 121 (40) (2024) e2412220121

    work page 2024

  28. [28]

    J. Liu, D. Chen, F. Pan, J. Wu, Neighborhood path complex for the quantitative analysis of the structure and stability of carboranes, Journal of Computational Biophysics and Chemistry 22 (04) (2023) 503–511

    work page 2023

  29. [29]

    Chowdhury, T

    S. Chowdhury, T. Gebhart, S. Huntsman, M. Yutin, Path homologies of deep feedforward networks, in: 2019 18th IEEE International Con- ference On Machine Learning And Applications (ICMLA), IEEE, 2019, pp. 1077–1082

    work page 2019

  30. [30]

    Chowdhury, S

    S. Chowdhury, S. Huntsman, M. Yutin, Path homology and temporal networks, in: International Conference on Complex Networks and Their Applications, Springer, 2020, pp. 639–650

    work page 2020

  31. [31]

    Edelsbrunner, J

    H. Edelsbrunner, J. Harer, et al., Persistent homology-a survey, Contemporary mathematics 453 (26) (2008) 257–282

    work page 2008

  32. [32]

    B. T. Fasy, F. Lecci, A. Rinaldo, L. Wasserman, S. Balakrishnan, A. Singh, Confidence sets for persistence diagrams (2014)

    work page 2014

  33. [33]

    D. E. Rumelhart, G. E. Hinton, R. J. Williams, Learning representa- tions by back-propagating errors, nature 323 (6088) (1986) 533–536

    work page 1986

  34. [34]

    Jaeger, Short term memory in echo state networks (2001)

    H. Jaeger, Short term memory in echo state networks (2001)

    work page 2001

  35. [35]

    Farkaš, R

    I. Farkaš, R. Bosák, P. Gergel’, Computational analysis of memory capacity in echo state networks, Neural Networks 83 (2016) 109–120

    work page 2016

  36. [36]

    Baran ˇcok, I

    P. Baran ˇcok, I. Farkaš, Memory capacity of input-driven echo state networks at the edge of chaos, in: Artificial Neural Networks and Machine Learning–ICANN 2014: 24th International Conference on Artificial Neural Networks, Hamburg, Germany, September 15-19,

    work page 2014

  37. [37]

    Proceedings 24, Springer, 2014, pp. 41–48

    work page 2014

  38. [38]

    Strauss, W

    T. Strauss, W. Wustlich, R. Labahn, Design strategies for weight matrices of echo state networks, Neural computation 24 (12) (2012) 3246–3276

    work page 2012

  39. [39]

    G. K. Venayagamoorthy, B. Shishir, Effects of spectral radius and settling time in the performance of echo state networks, Neural Networks 22 (7) (2009) 861–863

    work page 2009

  40. [40]

    C. Sun, M. Song, S. Hong, H. Li, A review of designs and ap- plications of echo state networks, arXiv preprint arXiv:2012.02974 (2020)

    work page arXiv 2012

  41. [41]

    F. Takens, Detecting strange attractors in turbulence, in: Dynamical Systems and Turbulence, Warwick 1980: proceedings of a sympo- sium held at the University of Warwick 1979/80, Springer, 2006, pp. 366–381

    work page 1980

  42. [42]

    Edelsbrunner, J

    H. Edelsbrunner, J. Harer, Computational topology: an introduction, American Mathematical Soc., 2010

    work page 2010

  43. [43]

    D. J. Watts, S. H. Strogatz, Collective dynamics of ‘small- world’networks, nature 393 (6684) (1998) 440–442

    work page 1998

  44. [44]

    Barabási, R

    A.-L. Barabási, R. Albert, Emergence of scaling in random networks, science 286 (5439) (1999) 509–512

    work page 1999

  45. [45]

    Z. Deng, Y . Zhang, Complex systems modeling using scale-free highly-clustered echo state network, in: The 2006 IEEE international joint conference on neural network proceedings, IEEE, 2006, pp. 3128–3135

    work page 2006

  46. [46]

    Kawai, J

    Y . Kawai, J. Park, M. Asada, A small-world topology enhances the echo state property and signal propagation in reservoir computing, Neural Networks 112 (2019) 15–23. :Preprint submitted to Elsevier Page 16 of 16

    work page 2019