pith. sign in

arxiv: 2607.01746 · v1 · pith:H3HZ7EDYnew · submitted 2026-07-02 · 💻 cs.LG

Finite-Lag Operator Geometry of Recurrent Representations

Kanishka Reddy This is my paper

Pith reviewed 2026-07-03 17:39 UTC · model grok-4.3

classification 💻 cs.LG
keywords finite-lag operator geometryrecurrent representationsconditional transport lawsource-centered transport tensorcoordinate circulationdeterministic recurrent motioncarre-du-champ geometrydense Gaussian estimator
0
0 comments X

The pith

A finite-lag conditional transport law from source-successor pairs decomposes recurrent dynamics into conditional spread and coherent displacement plus directed circulation.

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

The paper constructs a geometry for recurrent hidden states that treats them as trajectories observed at finite time lags rather than as static points. It begins with the conditional transport law Q_Δ(dy|x) between a source state and its successor after lag Δ, which is estimated by a dense Gaussian smoothing operator on observed pairs. From this law the authors derive a source-centered transport tensor G_Δ that splits exactly into a spread component and a coherent displacement component, together with an antisymmetric circulation tensor W_Δ^ρ that records net directed flow. They establish affine covariance of the derived quantities, stability of the dense estimator on bounded trajectory clouds, and a separation theorem showing that the finite-lag objects detect deterministic recurrent motion invisible to infinitesimal carre-du-champ geometry. The linear-Gaussian case supplies an explicit calibration in terms of the update matrix and covariances, and controlled experiments confirm the decomposition and the architecture-dependent differences it reveals in repeat-copy networks.

Core claim

From the directed finite-lag law we derive a source-centered transport tensor G_Δ, which decomposes exactly into conditional spread and coherent displacement, and an antisymmetric coordinate circulation W_Δ^ρ, which summarizes directed lagged flow. We prove affine covariance with explicit metric dependence of scalar summaries, dense estimator stability on bounded trajectory clouds, and a finite-lag separation result showing that source-centered transport detects deterministic recurrent motion not recorded by infinitesimal carre-du-champ geometry.

What carries the argument

The conditional transport law Q_Δ(dy|x) estimated by a dense Gaussian source-smoothing operator, which produces the source-centered transport tensor G_Δ and the antisymmetric circulation W_Δ^ρ.

If this is right

  • G_Δ is affine covariant and its scalar summaries depend explicitly on the chosen metric.
  • The dense estimator for G_Δ and W_Δ^ρ remains stable whenever trajectories remain inside a bounded cloud.
  • Source-centered transport separates deterministic recurrent motion from noise in a way infinitesimal geometry cannot.
  • In the linear-Gaussian case the quantities reduce to closed-form expressions involving the update matrix A_Δ, source covariance, and innovation covariance.
  • Architecture-dependent differences appear in total transport scale and coherent displacement trace when the same task is solved by different repeat-copy networks.

Where Pith is reading between the lines

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

  • The decomposition could be used to compare internal flow structure across recurrent architectures even when task performance is matched.
  • Because the circulation term is antisymmetric, it may serve as a diagnostic for directed information flow that is invisible to symmetric distance-based measures.
  • Metric dependence of the scalar summaries suggests that practitioners should select the embedding distance according to the physical or representational scale they wish to emphasize.
  • The finite-lag separation result raises the question of whether similar lag-based operators can be defined for non-Euclidean state spaces common in modern sequence models.

Load-bearing premise

The conditional transport law can be reliably estimated from observed source-successor pairs by a dense Gaussian source-smoothing operator, which requires bounded trajectory clouds.

What would settle it

In a linear-Gaussian recurrent system engineered to contain deterministic periodic orbits, measure whether the coherent-displacement trace of G_Δ is detectably positive while the corresponding carre-du-champ quadratic form remains zero.

Figures

Figures reproduced from arXiv: 2607.01746 by Kanishka Reddy.

Figure 1
Figure 1. Figure 1: Controlled calibration and stability of the finite-lag observables. (A) At fixed innovation [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Phase-resolved finite-lag geometry on repeat-copy with delay [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Controlled circulation sweep. The Frobenius norm of coordinate circulation increases with [PITH_FULL_IMAGE:figures/full_fig_p035_3.png] view at source ↗
read the original abstract

Recurrent representations are trajectories, but representation geometry is often measured from static snapshots. We develop finite-lag operator geometry for recurrent hidden states from observed source-successor pairs $(X_t,X_{t+\Delta})$. The primitive is the conditional transport law $Q_\Delta(dy\mid x)$, estimated by a dense Gaussian source-smoothing operator. From this directed finite-lag law we derive a source-centered transport tensor $G_\Delta$, which decomposes exactly into conditional spread and coherent displacement, and an antisymmetric coordinate circulation $W_\Delta^\rho$, which summarizes directed lagged flow. We prove affine covariance with explicit metric dependence of scalar summaries, dense estimator stability on bounded trajectory clouds, and a finite-lag separation result showing that source-centered transport detects deterministic recurrent motion not recorded by infinitesimal carre-du-champ geometry. A linear-Gaussian closed form calibrates the quantities in terms of the update $A_\Delta$, source covariance, and innovation covariance. Controlled experiments validate the decomposition, circulation, covariance, and stability predictions. In performance matched repeat-copy networks, the framework reveals architecture dependent differences in total transport scale and coherent displacement trace, while coherent displacement fraction is metric and resolution dependent.

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

0 major / 1 minor

Summary. The paper claims to develop finite-lag operator geometry for recurrent hidden states from observed source-successor pairs (X_t, X_{t+Δ}). The primitive is the conditional transport law Q_Δ(dy|x) estimated by a dense Gaussian source-smoothing operator. From this it derives the source-centered transport tensor G_Δ, which decomposes exactly into conditional spread and coherent displacement, and the antisymmetric coordinate circulation W_Δ^ρ summarizing directed lagged flow. It proves affine covariance with explicit metric dependence of scalar summaries, dense estimator stability on bounded trajectory clouds, and a finite-lag separation result showing source-centered transport detects deterministic recurrent motion not recorded by infinitesimal carre-du-champ geometry. A linear-Gaussian closed form calibrates the quantities in terms of the update A_Δ, source covariance, and innovation covariance. Controlled experiments validate the decomposition, circulation, covariance, and stability predictions, and the framework is applied to performance-matched repeat-copy networks.

Significance. If the derivations hold, the work supplies a new geometric framework for analyzing directed lagged flows in recurrent representations, with an explicit decomposition of transport and a proved distinction from existing infinitesimal geometry. The linear-Gaussian closed form and the controlled experiments validating multiple predictions are concrete strengths that support calibration and empirical checks.

minor comments (1)
  1. The description of the dense Gaussian source-smoothing operator used to estimate Q_Δ(dy|x) would benefit from an explicit statement of the bandwidth selection procedure and its sensitivity, as this directly affects the practical estimator whose stability is claimed.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the detailed and positive assessment of the manuscript, including the recognition of the derivations, the linear-Gaussian closed form, the controlled experiments, and the distinction from infinitesimal geometry. The recommendation for minor revision is appreciated. No specific major comments were listed in the report.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The derivation begins from the externally estimated conditional transport law Q_Δ(dy|x) obtained from observed source-successor pairs via a dense Gaussian source-smoothing operator. All subsequent objects (G_Δ, W_Δ^ρ, scalar summaries, affine covariance, stability bounds, and the finite-lag separation from carre-du-champ geometry) are obtained by explicit algebraic decomposition or proved consequences of this law under the stated bounded-trajectory-cloud assumption. The linear-Gaussian closed form is a direct specialization of the same transport law rather than a fitted parameter renamed as a prediction. No self-citation chain, self-definitional loop, or ansatz smuggled via prior work is present; the central claims remain independent of the target results and are externally falsifiable on the observed pairs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 2 invented entities

Based solely on the abstract, the framework rests on the Gaussian smoothing estimator and bounded-cloud stability; new tensors and circulation are introduced without independent evidence outside the derivation.

axioms (2)
  • standard math Affine covariance of scalar summaries with explicit metric dependence
    Stated as proved in the paper for the transport tensor
  • domain assumption Dense estimator stability on bounded trajectory clouds
    Invoked to support the Gaussian source-smoothing operator
invented entities (2)
  • source-centered transport tensor G_Δ no independent evidence
    purpose: Decomposes conditional transport into spread and coherent displacement
    New object derived from the finite-lag law
  • antisymmetric coordinate circulation W_Δ^ρ no independent evidence
    purpose: Summarizes directed lagged flow
    New object derived from the finite-lag law

pith-pipeline@v0.9.1-grok · 5727 in / 1389 out tokens · 27298 ms · 2026-07-03T17:39:21.981832+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

49 extracted references · 28 canonical work pages

  1. [1]

    SVCCA: Singu- lar vector canonical correlation analysis for deep learning dynamics and interpretability

    Maithra Raghu, Justin Gilmer, Jason Yosinski, and Jascha Sohl-Dickstein. SVCCA: Singu- lar vector canonical correlation analysis for deep learning dynamics and interpretability. In Advances in Neural Information Processing Systems, volume 30, pages 6076–6085, 2017

    2017

  2. [2]

    Similarity of neural network representations revisited

    Simon Kornblith, Mohammad Norouzi, Honglak Lee, and Geoffrey Hinton. Similarity of neural network representations revisited. InProceedings of the 36th International Conference on Machine Learning, volume 97 ofProceedings of Machine Learning Research, pages 3519–3529. PMLR, 2019

    2019

  3. [3]

    Macke, and Davide Zoccolan

    Alessio Ansuini, Alessandro Laio, Jakob H. Macke, and Davide Zoccolan. Intrinsic dimension of data representations in deep neural networks. InAdvances in Neural Information Processing Systems, volume 32, 2019

    2019

  4. [4]

    Lee, and Haim Sompolinsky

    Uri Cohen, SueYeon Chung, Daniel D. Lee, and Haim Sompolinsky. Separability and geometry of object manifolds in deep neural networks.Nature Communications, 11(1):746, 2020. doi: 10.1038/s41467-020-14578-5

    work page doi:10.1038/s41467-020-14578-5 2020

  5. [5]

    Vardan Papyan, X. Y . Han, and David L. Donoho. Prevalence of neural collapse during the terminal phase of deep learning training.Proceedings of the National Academy of Sciences, 117 (40):24652–24663, 2020. doi: 10.1073/pnas.2015509117

    work page doi:10.1073/pnas.2015509117 2020

  6. [6]

    Christensen, Alexander Tong, Guillaume Huguet, Guy Wolf, Maximilian Nickel, Ian Adelstein, and Smita Krishnaswamy

    Danqi Liao, Chen Liu, Benjamin W. Christensen, Alexander Tong, Guillaume Huguet, Guy Wolf, Maximilian Nickel, Ian Adelstein, and Smita Krishnaswamy. Assessing neural network representations during training using noise-resilient diffusion spectral entropy, 2023

    2023

  7. [7]

    Steindl, Selma Mazioud, Ellie Schueler, Folu Ogundipe, Ellen Zhang, Yvan Grinspan, Kristof Reimann, Peyton Crevasse, Dhananjay Bhaskar, Siddharth Viswanath, Yanlei Zhang, Tim G

    Elliott Abel, Andrew J. Steindl, Selma Mazioud, Ellie Schueler, Folu Ogundipe, Ellen Zhang, Yvan Grinspan, Kristof Reimann, Peyton Crevasse, Dhananjay Bhaskar, Siddharth Viswanath, Yanlei Zhang, Tim G. J. Rudner, Ian Adelstein, and Smita Krishnaswamy. Exploring the manifold of neural networks using diffusion geometry, 2024

    2024

  8. [8]

    Laplacian eigenmaps for dimensionality reduction and data representation.Neural Computation, 15(6):1373–1396, 2003

    Mikhail Belkin and Partha Niyogi. Laplacian eigenmaps for dimensionality reduction and data representation.Neural Computation, 15(6):1373–1396, 2003. doi: 10.1162/ 089976603321780317

    2003

  9. [9]

    Coifman and Stéphane Lafon

    Ronald R. Coifman and Stéphane Lafon. Diffusion maps.Applied and Computational Harmonic Analysis, 21(1):5–30, 2006. doi: 10.1016/j.acha.2006.04.006

    work page doi:10.1016/j.acha.2006.04.006 2006

  10. [10]

    Diffusion operator geometry of feedforward representations, 2026

    Kanishka Reddy. Diffusion operator geometry of feedforward representations, 2026

    2026

  11. [11]

    Peter J. Schmid. Dynamic mode decomposition of numerical and experimental data.Journal of Fluid Mechanics, 656:5–28, 2010. doi: 10.1017/S0022112010001217

    work page doi:10.1017/s0022112010001217 2010

  12. [12]

    Williams, Ioannis G

    Matthew O. Williams, Ioannis G. Kevrekidis, and Clarence W. Rowley. A data-driven ap- proximation of the koopman operator: Extending dynamic mode decomposition.Journal of Nonlinear Science, 25(6):1307–1346, 2015. doi: 10.1007/s00332-015-9258-5

    work page doi:10.1007/s00332-015-9258-5 2015

  13. [13]

    Identification of slow molecular order parameters for markov model construction.The Journal of Chemical Physics, 139(1):015102, 2013

    Guillermo Pérez-Hernández, Fabian Paul, Toni Giorgino, Gianni De Fabritiis, and Frank Noé. Identification of slow molecular order parameters for markov model construction.The Journal of Chemical Physics, 139(1):015102, 2013. doi: 10.1063/1.4811489

    work page doi:10.1063/1.4811489 2013

  14. [14]

    Schwantes and Vijay S

    Christian R. Schwantes and Vijay S. Pande. Improvements in markov state model construction reveal many non-native interactions in the folding of ntl9.Journal of Chemical Theory and Computation, 9(4):2000–2009, 2013. doi: 10.1021/ct300878a

    work page doi:10.1021/ct300878a 2000

  15. [15]

    Variational approach for learning markov processes from time series data.Journal of Nonlinear Science, 30(1):23–66, 2020

    Hao Wu and Frank Noé. Variational approach for learning markov processes from time series data.Journal of Nonlinear Science, 30(1):23–66, 2020. doi: 10.1007/s00332-019-09567-y

    work page doi:10.1007/s00332-019-09567-y 2020

  16. [16]

    V AMPnets for deep learning of molecular kinetics.Nature Communications, 9(1):5, 2018

    Andreas Mardt, Luca Pasquali, Hao Wu, and Frank Noé. V AMPnets for deep learning of molecular kinetics.Nature Communications, 9(1):5, 2018. doi: 10.1038/s41467-017-02388-1

    work page doi:10.1038/s41467-017-02388-1 2018

  17. [17]

    Diffusions hypercontractives.Séminaire de Probabilités XIX 1983/84, 1123:177–206, 1985

    Dominique Bakry and Michel Émery. Diffusions hypercontractives.Séminaire de Probabilités XIX 1983/84, 1123:177–206, 1985. doi: 10.1007/BFb0075847. 10

    work page doi:10.1007/bfb0075847 1983

  18. [18]

    Springer, 2014

    Dominique Bakry, Ivan Gentil, and Michel Ledoux.Analysis and Geometry of Markov Diffusion Operators, volume 348 ofGrundlehren der mathematischen Wissenschaften. Springer, 2014. doi: 10.1007/978-3-319-00227-9

    work page doi:10.1007/978-3-319-00227-9 2014

  19. [19]

    Coifman, and Ioannis G

    Boaz Nadler, Stephane Lafon, Ronald R. Coifman, and Ioannis G. Kevrekidis. Diffusion maps, spectral clustering and reaction coordinates of dynamical systems.Applied and Computational Harmonic Analysis, 21(1):113–127, 2006. doi: 10.1016/j.acha.2005.07.004

    work page doi:10.1016/j.acha.2005.07.004 2006

  20. [20]

    Diffusion geometry, 2024

    Iolo Jones. Diffusion geometry, 2024

    2024

  21. [21]

    Manifold diffusion geometry: Curvature, tangent spaces, and dimension, 2024

    Iolo Jones. Manifold diffusion geometry: Curvature, tangent spaces, and dimension, 2024

    2024

  22. [22]

    Computing diffusion geometry, 2026

    Iolo Jones and David Lanners. Computing diffusion geometry, 2026

    2026

  23. [23]

    Williams, Clarence W

    Matthew O. Williams, Clarence W. Rowley, and Ioannis G. Kevrekidis. A kernel-based method for data-driven koopman spectral analysis.Journal of Computational Dynamics, 2(2):247–265,

  24. [24]

    doi: 10.3934/jcd.2015005

    work page doi:10.3934/jcd.2015005

  25. [25]

    Data-driven model reduction and transfer operator approximation.Journal of Nonlinear Science, 28(3):985–1010, 2018

    Stefan Klus, Feliks Nuske, Péter Koltai, Hao Wu, Ioannis Kevrekidis, Christof Schutte, and Frank Noé. Data-driven model reduction and transfer operator approximation.Journal of Nonlinear Science, 28(3):985–1010, 2018. doi: 10.1007/s00332-017-9437-7

    work page doi:10.1007/s00332-017-9437-7 2018

  26. [26]

    Eigendecompositions of transfer opera- tors in reproducing kernel hilbert spaces.Journal of Nonlinear Science, 30:283–315, 2020

    Stefan Klus, Ingmar Schuster, and Krikamol Muandet. Eigendecompositions of transfer opera- tors in reproducing kernel hilbert spaces.Journal of Nonlinear Science, 30:283–315, 2020. doi: 10.1007/s00332-019-09574-z

    work page doi:10.1007/s00332-019-09574-z 2020

  27. [27]

    An analytic framework for identifying finite-time coherent sets in time- dependent dynamical systems.Physica D: Nonlinear Phenomena, 250:1–19, 2013

    Gary Froyland. An analytic framework for identifying finite-time coherent sets in time- dependent dynamical systems.Physica D: Nonlinear Phenomena, 250:1–19, 2013. doi: 10.1016/j.physd.2013.01.013

    work page doi:10.1016/j.physd.2013.01.013 2013

  28. [28]

    Dynamic isoperimetry and the geometry of lagrangian coherent structures

    Gary Froyland. Dynamic isoperimetry and the geometry of lagrangian coherent structures. Nonlinearity, 28(10):3587–3622, 2015. doi: 10.1088/0951-7715/28/10/3587

    work page doi:10.1088/0951-7715/28/10/3587 2015

  29. [29]

    Coifman and Matthew J

    Ronald R. Coifman and Matthew J. Hirn. Diffusion maps for changing data.Applied and Computational Harmonic Analysis, 36(1):79–107, 2014. doi: 10.1016/j.acha.2013.03.001

    work page doi:10.1016/j.acha.2013.03.001 2014

  30. [30]

    Marshall and Matthew J

    Nicholas F. Marshall and Matthew J. Hirn. Time coupled diffusion maps.Applied and Computational Harmonic Analysis, 45(3):709–728, 2018. doi: 10.1016/j.acha.2017.08.007

    work page doi:10.1016/j.acha.2017.08.007 2018

  31. [31]

    Opening the black box: Low-dimensional dynamics in high- dimensional recurrent neural networks

    David Sussillo and Omri Barak. Opening the black box: Low-dimensional dynamics in high- dimensional recurrent neural networks. InAdvances in Neural Information Processing Systems, volume 26, 2013

    2013

  32. [32]

    Neural circuits as computational dynamical systems.Current Opinion in Neurobiology, 25:156–163, 2014

    David Sussillo. Neural circuits as computational dynamical systems.Current Opinion in Neurobiology, 25:156–163, 2014. doi: 10.1016/j.conb.2014.01.008

    work page doi:10.1016/j.conb.2014.01.008 2014

  33. [33]

    McIntosh, David B

    Niru Maheswaranathan, Lane T. McIntosh, David B. Kastner, Josh Melander, L. E. Brezovec, Aran Nayebi, Julia Wang, Surya Ganguli, and Stephen A. Baccus. Reverse engineering recurrent networks for sentiment classification reveals line attractor dynamics, 2019

    2019

  34. [34]

    Williams, Matthew D

    Niru Maheswaranathan, Alex H. Williams, Matthew D. Golub, Surya Ganguli, and David Sussillo. Universality and individuality in neural dynamics across large populations of recurrent networks. InAdvances in Neural Information Processing Systems, volume 32, 2019

    2019

  35. [35]

    Rylan Schaeffer, Mikail Khona, Leenoy Meshulam, International Brain Laboratory, and Ila R. Fiete. Reverse-engineering recurrent neural network solutions to a hierarchical inference task for mice. InAdvances in Neural Information Processing Systems, volume 33, pages 4584–4596, 2020

    2020

  36. [36]

    Jimmy T. H. Smith, Scott W. Linderman, and David Sussillo. Reverse engineering recurrent neural networks with jacobian switching linear dynamical systems. InAdvances in Neural Information Processing Systems, volume 34, pages 16700–16713, 2021. 11

    2021

  37. [37]

    Linking connectivity, dynamics, and com- putations in low-rank recurrent neural networks.Neuron, 99(3):609–623.e29, 2018

    Francesca Mastrogiuseppe and Srdjan Ostojic. Linking connectivity, dynamics, and com- putations in low-rank recurrent neural networks.Neuron, 99(3):609–623.e29, 2018. doi: 10.1016/j.neuron.2018.07.003

    work page doi:10.1016/j.neuron.2018.07.003 2018

  38. [38]

    Shaping dynamics with multiple populations in low-rank recurrent networks.Neural Computation, 33(6):1572–1615, 2021

    Manuel Beiran, Alexis Dubreuil, Adrian Valente, Francesca Mastrogiuseppe, and Srdjan Os- tojic. Shaping dynamics with multiple populations in low-rank recurrent networks.Neural Computation, 33(6):1572–1615, 2021. doi: 10.1162/neco_a_01381

    work page doi:10.1162/neco_a_01381 2021

  39. [39]

    Pillow, and Srdjan Ostojic

    Adrian Valente, Jonathan W. Pillow, and Srdjan Ostojic. Extracting computational mechanisms from neural data using low-rank RNNs. InAdvances in Neural Information Processing Systems, volume 35, pages 24072–24086, 2022

    2022

  40. [40]

    Macke, and Omri Barak

    Matthijs Pals, Jakob H. Macke, and Omri Barak. Trained recurrent neural networks develop phase-locked limit cycles in a working memory task.PLOS Computational Biology, 20(2): e1011852, 2024. doi: 10.1371/journal.pcbi.1011852

    work page doi:10.1371/journal.pcbi.1011852 2024

  41. [41]

    Recurrent neural networks with transient trajectory explain working memory en- coding mechanisms.Communications Biology, 8:88, 2025

    Chen Liu et al. Recurrent neural networks with transient trajectory explain working memory en- coding mechanisms.Communications Biology, 8:88, 2025. doi: 10.1038/s42003-024-07282-3

    work page doi:10.1038/s42003-024-07282-3 2025

  42. [42]

    Dynamical phases of short-term memory mechanisms in RNNs

    Bariscan Kurtkaya, Fatih Dinc, Mert Yuksekgonul, Marta Blanco-Pozo, Ege Cirakman, Mark Schnitzer, Yucel Yemez, Hidenori Tanaka, Peng Yuan, and Nina Miolane. Dynamical phases of short-term memory mechanisms in RNNs. InProceedings of the 42nd International Conference on Machine Learning, volume 267 ofProceedings of Machine Learning Research, pages 32032–320...

    2025

  43. [43]

    Sejnowski, and Hava T

    Arjun Karuvally, Terrence J. Sejnowski, and Hava T. Siegelmann. Hidden traveling waves bind working memory variables in recurrent neural networks. InProceedings of the 41st International Conference on Machine Learning, volume 235 ofProceedings of Machine Learning Research. PMLR, 2024

    2024

  44. [44]

    Driscoll, Krishna V

    Laura N. Driscoll, Krishna V . Shenoy, and David Sussillo. Flexible multitask computation in recurrent networks utilizes shared dynamical motifs.Nature Neuroscience, 27(7):1349–1363,

  45. [45]

    doi: 10.1038/s41593-024-01668-6

    work page doi:10.1038/s41593-024-01668-6

  46. [46]

    Long short-term memory.Neural Computation, 9 (8):1735–1780, 1997

    Sepp Hochreiter and Jurgen Schmidhuber. Long short-term memory.Neural Computation, 9 (8):1735–1780, 1997. doi: 10.1162/neco.1997.9.8.1735

    work page doi:10.1162/neco.1997.9.8.1735 1997

  47. [47]

    Learning phrase representations using RNN encoder– decoder for statistical machine translation

    Kyunghyun Cho, Bart van Merrienboer, Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. Learning phrase representations using RNN encoder– decoder for statistical machine translation. InProceedings of the 2014 Conference on Empirical Methods in Natural Language Processing, pages 1724–1734, 2014. doi: 10.3115/v1/D14-1179

    work page doi:10.3115/v1/d14-1179 2014

  48. [48]

    Lipschitz regularity of graph laplacians on random data clouds.SIAM Journal on Mathematical Analysis, 54(1):1169–1222, 2022

    Jeff Calder, Nicolás García Trillos, and Marta Lewicka. Lipschitz regularity of graph laplacians on random data clouds.SIAM Journal on Mathematical Analysis, 54(1):1169–1222, 2022. doi: 10.1137/20M1356610

    work page doi:10.1137/20m1356610 2022

  49. [49]

    the carré du champ on ht

    Daniel Ting, Ling Huang, and Michael I. Jordan. An analysis of the convergence of graph laplacians. InProceedings of the 27th International Conference on Machine Learning, pages 1079–1086, 2010. 12 A Empirical operator construction Here we give the operational details of the empirical finite-lag operatorbP∆ used throughout the paper. The dense Gaussian so...

    2010