pith. sign in

arxiv: 2508.18948 · v2 · submitted 2025-08-26 · ✦ hep-th · cond-mat.dis-nn· cs.LG· stat.ML

Gauge-covariant stochastic neural fields: Stability and finite-width effects

Rodrigo Carmo Terin This is my paper

Pith reviewed 2026-05-18 21:41 UTC · model grok-4.3

classification ✦ hep-th cond-mat.dis-nncs.LGstat.ML
keywords stochastic neural fieldsgauge covariancefinite-width effectsneural network stabilityLyapunov exponentedge of chaoseffective field theoryMartin-Siggia-Rose formalism
0
0 comments X

The pith

A gauge-covariant stochastic field theory shows finite-width effects correct neural kernels perturbatively without shifting the marginality condition.

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

The paper develops a gauge-covariant stochastic effective field theory to analyze stability and finite-width effects in deep neural systems. It introduces classical commuting fields consisting of a complex matter field, a real Abelian connection field, and a fictitious stochastic depth variable, then applies the Martin-Siggia-Rose-Janssen-de Dominicis formalism to obtain a functional representation and a two-replica linear-response construction for the maximal Lyapunov exponent and amplification factor. The central result is that finite-width effects enter as perturbative corrections to dressed kernels while the marginality condition for the edge of chaos remains unchanged at the order considered when kernel geometry is held fixed. Numerical tests confirm that finite-width multilayer perceptrons track the mean-field instability threshold and that a linear stochastic sector reproduces the expected low-frequency spectral deformation.

Core claim

The central claim is that a gauge-covariant stochastic effective field theory built from classical commuting fields—a complex matter field, a real Abelian connection field, and a fictitious stochastic depth variable—together with the Martin-Siggia-Rose-Janssen-de Dominicis formalism yields a functional representation and two-replica linear-response construction that defines the maximal Lyapunov exponent and amplification factor. Finite-width effects appear as perturbative corrections to dressed kernels, and the marginality condition remains unchanged at the order considered for fixed kernel geometry. This description is supported by the observation that finite-width multilayer perceptrons he

What carries the argument

The gauge-covariant stochastic effective field theory constructed via classical commuting fields and the Martin-Siggia-Rose-Janssen-de Dominicis formalism, which supplies the two-replica linear-response construction for Lyapunov exponents and amplification factors.

If this is right

  • Finite-width multilayer perceptrons follow the mean-field instability threshold.
  • The marginality condition for stability remains unchanged under the considered perturbative finite-width corrections for fixed kernel geometry.
  • A linear stochastic effective sector reproduces the predicted low-frequency spectral deformation.
  • Finite-width effects manifest specifically as perturbative corrections to the dressed kernels.

Where Pith is reading between the lines

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

  • The same construction could be used to track how learned or variable kernel geometries alter the marginality condition beyond the fixed-geometry case.
  • Extending the stochastic sector to include nonlinear terms might generate testable predictions for higher-order finite-width corrections to stability.
  • The gauge-covariant setup suggests a route to incorporate additional symmetries when modeling dynamics in other layered systems.

Load-bearing premise

The classical commuting fields for matter, connection, and stochastic depth together with the Martin-Siggia-Rose-Janssen-de Dominicis formalism provide an accurate effective description of the stochastic dynamics and stability properties in deep neural systems.

What would settle it

A numerical simulation of finite-width networks that shows the instability threshold or low-frequency spectrum deviating from the mean-field prediction and the linear-sector deformation in a manner not captured by the perturbative corrections to dressed kernels.

Figures

Figures reproduced from arXiv: 2508.18948 by Rodrigo Carmo Terin.

Figure 1
Figure 1. Figure 1: Empirical largest Lyapunov exponent per layer compared to the mean-field prediction [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Schematic of a U(1) GINN layer. Each neuron i at layer ℓ has an associated phase factor e iθ(ℓ) i , and each weight connecting neurons between layers carries a link variable A (ℓ) ij . This ensures gauge covariance of the update rule under local transformations. During training the magnitudes w (ℓ) ij and phases A (ℓ) ij are updated by gradient descent. Because only gauge–invariant combinations of these pa… view at source ↗
Figure 3
Figure 3. Figure 3: One-loop QED–NN diagrams consistent with Eqs. (C1)–(C2). [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
read the original abstract

We develop a gauge-covariant stochastic effective field theory for stability and finite-width effects in deep neural systems. The model uses classical commuting fields: a complex matter field, a real Abelian connection field, and a fictitious stochastic depth variable. Using the Martin--Siggia--Rose--Janssen--de~Dominicis formalism, we derive its functional representation and a two-replica linear-response construction defining the maximal Lyapunov exponent and the amplification factor for the edge of chaos. Finite-width effects appear as perturbative corrections to dressed kernels, and the marginality condition remains unchanged at the order considered for fixed kernel geometry. Numerically, finite-width multilayer perceptrons follow the mean-field instability threshold, and a linear stochastic effective sector reproduces the predicted low-frequency spectral deformation.

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 develops a gauge-covariant stochastic effective field theory for stability and finite-width effects in deep neural systems. It introduces classical commuting fields (a complex matter field, a real Abelian connection field, and a fictitious stochastic depth variable), applies the Martin-Siggia-Rose-Janssen-de Dominicis formalism to obtain a functional representation, and constructs a two-replica linear-response analysis to define the maximal Lyapunov exponent and amplification factor at the edge of chaos. The central claims are that finite-width effects enter as perturbative corrections to dressed kernels and that the marginality condition remains unchanged at the order considered when kernel geometry is held fixed; numerical support is provided by finite-width multilayer perceptrons tracking the mean-field instability threshold and by a linear stochastic effective sector reproducing predicted low-frequency spectral deformation.

Significance. If the derivations and numerics hold, the work supplies a field-theoretic language that incorporates gauge covariance and stochastic depth into neural-network stability analysis, with the result that mean-field marginality is robust against finite-width corrections at the perturbative order examined. The numerical reproduction of both the instability threshold and low-frequency spectral features provides concrete evidence that the effective description captures relevant dynamics, which could be useful for analytic control of finite-width effects in deep networks.

major comments (2)
  1. The claim that the marginality condition remains unchanged relies on treating kernel geometry as an external fixed input to the two-replica linear-response analysis. The manuscript does not derive or bound possible O(1/N) corrections to the effective kernel spectrum (eigenvalue distribution or singular values) arising from the gauge-covariant fields and stochastic depth variable; such corrections would back-react on the dressed propagators and could shift the location of the marginality condition, rendering the 'unchanged' statement order-dependent rather than robust.
  2. The abstract states that finite-width multilayer perceptrons follow the mean-field instability threshold and that the linear stochastic sector reproduces the predicted low-frequency spectral deformation, yet no explicit equations, error bars, data-exclusion criteria, or quantitative measures of agreement (e.g., deviation from the mean-field threshold) are provided in the summary. Without these details the numerical support for the central claim cannot be independently verified.
minor comments (2)
  1. Notation for the fictitious stochastic depth variable and its coupling to the connection field should be introduced with an explicit definition and a clear statement of its physical interpretation before the MSRJD construction is applied.
  2. The manuscript would benefit from a short table or figure summarizing the perturbative orders at which kernel corrections are computed versus the orders at which they are neglected.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading and constructive comments. We address the two major comments point by point below, indicating where revisions will be made to the manuscript.

read point-by-point responses

  1. Referee: The claim that the marginality condition remains unchanged relies on treating kernel geometry as an external fixed input to the two-replica linear-response analysis. The manuscript does not derive or bound possible O(1/N) corrections to the effective kernel spectrum (eigenvalue distribution or singular values) arising from the gauge-covariant fields and stochastic depth variable; such corrections would back-react on the dressed propagators and could shift the location of the marginality condition, rendering the 'unchanged' statement order-dependent rather than robust.

    Authors: We thank the referee for this observation. Our analysis is performed at a fixed perturbative order in which the kernel geometry is treated as an external input parameter, as is standard when isolating finite-width corrections to the dressed propagators. At this order the marginality condition is indeed unchanged. We acknowledge that a self-consistent determination of O(1/N) corrections to the kernel spectrum itself lies beyond the present scope and could in principle induce back-reaction. We will revise the manuscript to state this limitation explicitly and to note that a full treatment of kernel-spectrum corrections remains an open question for future work. revision: partial

  2. Referee: The abstract states that finite-width multilayer perceptrons follow the mean-field instability threshold and that the linear stochastic sector reproduces the predicted low-frequency spectral deformation, yet no explicit equations, error bars, data-exclusion criteria, or quantitative measures of agreement (e.g., deviation from the mean-field threshold) are provided in the summary. Without these details the numerical support for the central claim cannot be independently verified.

    Authors: We agree that additional quantitative details will improve verifiability. In the revised manuscript we will supply the explicit equations used to compute the instability threshold and spectral deformation, include error bars on all numerical curves, state any data-exclusion criteria employed, and report quantitative measures such as the mean absolute deviation from the mean-field threshold. These elements are already contained in our numerical analysis and will be clearly presented in the main text. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation remains self-contained

full rationale

The paper constructs a gauge-covariant stochastic EFT from classical commuting fields (complex matter, real Abelian connection, fictitious stochastic depth) and applies the standard MSRJD formalism to obtain a two-replica linear-response analysis for the maximal Lyapunov exponent and edge-of-chaos amplification factor. Finite-width corrections are derived perturbatively on dressed kernels while holding kernel geometry fixed at the order considered; the unchanged marginality condition is presented as a direct consequence of that perturbative expansion rather than a redefinition or fit of the input geometry itself. Numerical checks on finite-width MLPs and a linear stochastic sector are reported as independent verification against the mean-field threshold and low-frequency spectral deformation. No self-definitional reductions, fitted inputs relabeled as predictions, or load-bearing self-citations appear in the derivation chain. The framework is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 3 invented entities

The model rests on domain assumptions from statistical field theory and introduces several new entities without independent evidence outside the construction; no explicit free parameters are named in the abstract but the perturbative treatment implies choices in kernel geometry and expansion order.

axioms (1)
  • domain assumption The Martin-Siggia-Rose-Janssen-de Dominicis formalism applies directly to the stochastic dynamics of neural networks.
    Invoked to derive the functional representation and linear-response construction.
invented entities (3)
  • Complex matter field no independent evidence
    purpose: Represent neural activations in a gauge-covariant manner.
    Introduced as a core component of the effective field theory.
  • Real Abelian connection field no independent evidence
    purpose: Enforce gauge covariance in the stochastic description.
    New field postulated to achieve the gauge-covariant property.
  • Fictitious stochastic depth variable no independent evidence
    purpose: Model the stochastic progression through network depth.
    Invented entity to incorporate depth into the field theory.

pith-pipeline@v0.9.0 · 5660 in / 1555 out tokens · 53412 ms · 2026-05-18T21:41:58.091855+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

80 extracted references · 80 canonical work pages · 2 internal anchors

  1. [1]

    Imagenet classification with deep convolutional neural networks

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. In Advances in Neural Information Processing Systems 25 , pages 1097–1105, 2012

    work page 2012

  2. [2]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pages 770–778, 2016

    work page 2016

  3. [3]

    Deep Speech: Scaling up end-to-end speech recognition

    Awni Hannun, Carl Case, Jared Casper, Bryan Catanzaro, Greg Diamos, Erich Elsen, Ryan Prenger, Sanjeev Satheesh, Shubho Sengupta, Adam Coates, Andrew Y Ng, et al. Deep speech: Scaling up end-to-end speech recognition. arXiv preprint arXiv:1412.5567, 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  4. [4]

    Long short-term memory recurrent neural network architectures for large scale acoustic modeling

    Hasim Sak, Andrew Senior, and Fran¸ coise Beaufays. Long short-term memory recurrent neural network architectures for large scale acoustic modeling. In Proceedings of Interspeech 2014, pages 338–342. ISCA, 2014

    work page 2014

  5. [5]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in Neural Information Processing Systems , pages 5998–6008, 2017

    work page 2017

  6. [6]

    Bert: Pre-training of deep bidirectional transformers for language understanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) , pages 4171–4186. Associatio...

    work page 2019

  7. [7]

    Ai researchers allege that machine learning is alchemy

    Matthew Hutson. Ai researchers allege that machine learning is alchemy. Science Magazine, 2018

    work page 2018

  8. [8]

    Why is ai hard and physics simple? arXiv preprint arXiv:2104.00008 , 2021

    Daniel A Roberts. Why is ai hard and physics simple? arXiv preprint arXiv:2104.00008 , 2021

    work page arXiv 2021

  9. [9]

    Non-gaussian processes and neural networks at finite widths

    Sho Yaida. Non-gaussian processes and neural networks at finite widths. In Jianfeng Lu and Rachel Ward, editors, Proceedings of The First Mathematical and Scientific Machine Learning Conference, volume 107 of Proceedings of Machine Learning Research, pages 165–192. PMLR, 20–24 Jul 2020

    work page 2020

  10. [10]

    Asymptotics of wide networks from feynman diagrams

    Ethan Dyer and Guy Gur-Ari. Asymptotics of wide networks from feynman diagrams. In Eighth International Conference on Learning Representations, 2020

    work page 2020

  11. [11]

    Neural networks and quantum field theory

    James Halverson, Anirudh Maiti, and Keegan Stoner. Neural networks and quantum field theory. Machine Learning: Science and Technology, 2(3):035002, 2021

    work page 2021

  12. [12]

    Nonperturbative renormalization for the neural network–qft correspondence

    Harold Erbin, Vincent Lahoche, and Dine Ousmane Samary. Nonperturbative renormalization for the neural network–qft correspondence. Machine Learning: Science and Technology , 3(1):015027, 2022. 17

    work page 2022

  13. [13]

    Symmetry-via-duality: invariant neural network densities from parameter-space correlators

    Anirudh Maiti, Keegan Stoner, and James Halverson. Symmetry-via-duality: invariant neural network densities from parameter-space correlators. arXiv preprint arXiv:2106.00694 , 2021

    work page arXiv 2021

  14. [14]

    Grosvenor and Ro Jefferson

    Kevin T. Grosvenor and Ro Jefferson. The edge of chaos: quantum field theory and deep neural networks. SciPost Physics, 12(3):081, 2022

    work page 2022

  15. [15]

    Bayesian learning for neural networks

    Radford M Neal. Bayesian learning for neural networks . PhD thesis, University of Toronto, 1995

    work page 1995

  16. [16]

    Deep Neural Networks as Gaussian Processes

    Jaehoon Lee, Yasaman Bahri, Roman Novak, Samuel S Schoenholz, Jeffrey Pennington, and Jascha Sohl-Dickstein. Deep neural networks as gaussian processes. arXiv preprint arXiv:1711.00165 , 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  17. [17]

    The principles of deep learning theory

    Daniel A Roberts, Sho Yaida, and Boris Hanin. The principles of deep learning theory . Cambridge University Press, 2022

    work page 2022

  18. [18]

    Critical behavior of a classical heisenberg ferromagnet with many degrees of freedom

    Edouard Br´ ezin and David J Wallace. Critical behavior of a classical heisenberg ferromagnet with many degrees of freedom. Physical Review B, 7(5):1967, 1973

    work page 1967

  19. [19]

    Quantum field theory in the large N limit: a review , volume 385

    Moshe Moshe and Jean Zinn-Justin. Quantum field theory in the large N limit: a review , volume 385. Elsevier, 2003

    work page 2003

  20. [20]

    The hintons in your neural network: a quantum field theory view of deep learning

    Roberto Bondesan and Max Welling. The hintons in your neural network: a quantum field theory view of deep learning. In Marina Meila and Tong Zhang, editors, Proceedings of the 38th International Conference on Machine Learning , volume 139 of Proceedings of Machine Learning Research, pages 1038–1048. PMLR, 18–24 Jul 2021

    work page 2021

  21. [21]

    P. A. M. Dirac. The quantum theory of the emission and absorption of radiation. Proceedings of the Royal Society A , 114(767):243–265, 1927

    work page 1927

  22. [22]

    M. E. Peskin and D. V. Schroeder. An Introduction to Quantum Field Theory . CRC Press, 2018

    work page 2018

  23. [23]

    W. J. Marciano and H. Pagels. Quantum electrodynamics. Physics Reports, 36(3):137–276, 1978

    work page 1978

  24. [24]

    J. S. Ball and T. W. Chiu. Analytic properties of the vertex function in gauge theories. i. Physical Review D, 22(10):2542– 2549, 1980

    work page 1980

  25. [25]

    Karplus and M

    R. Karplus and M. Neuman. The scattering of light by light. Physical Review, 80(2):380–385, 1950

    work page 1950

  26. [26]

    Karplus and M

    R. Karplus and M. Neuman. Non-linear interactions between electromagnetic fields. Physical Review, 83(4):776–784, 1951

    work page 1951

  27. [27]

    L. C. Loveridge, O. Oliveira, and P. J. Silva. Lattice study of the photon propagator in qed. Physical Review D , 103(9):094519, 2021

    work page 2021

  28. [28]

    L. C. Loveridge, O. Oliveira, and P. J. Silva. Landau gauge in compact lattice qed. Physical Review D , 104(11):114511, 2021

    work page 2021

  29. [29]

    L. C. Loveridge, O. Oliveira, and P. J. Silva. Infrared properties of the qed photon propagator. Physical Review D , 106(1):L011502, 2022

    work page 2022

  30. [30]

    Oliveira, P

    O. Oliveira, P. J. Silva, J. I. Skullerud, and A. Sternbeck. Photon propagator in lattice qed.Physical Review D, 99(9):094506, 2019

    work page 2019

  31. [31]

    Kızılers¨ u, T

    A. Kızılers¨ u, T. Sizer, M. R. Pennington, A. G. Williams, and R. Williams. Nonperturbative studies of the fermion-photon vertex. Physical Review D, 91(6):065015, 2015

    work page 2015

  32. [32]

    Bashir, A

    A. Bashir, A. Raya, et al. Gauge covariance of the fermion-boson vertex in qed. Physical Review D, 89(1):016014, 2014

    work page 2014

  33. [33]

    H. Gies. Introduction to the functional rg and applications to gauge theories. Lecture Notes in Physics, 852:287–348, 2012

    work page 2012

  34. [34]

    H. Pagels. Dynamical chiral symmetry breaking in qed. Physical Review D, 19(11):3080–3088, 1979

    work page 1979

  35. [35]

    Atkinson, P

    D. Atkinson, P. W. Johnson, and P. Maris. Dynamical mass generation in qed. Physical Review D, 42(2):602–609, 1990

    work page 1990

  36. [36]

    Rojas, A

    E. Rojas, A. Ayala, A. Bashir, and A. Raya. Dynamical mass generation in quenched qed. Physical Review D, 77(9):093004, 2008

    work page 2008

  37. [37]

    C. S. Fischer, R. Alkofer, T. Dahm, and P. Maris. Dynamical chiral symmetry breaking in unquenched qed. Physical Review D, 70(7):073007, 2004

    work page 2004

  38. [38]

    J. M. Cornwall. Dynamical mass generation in continuum qed. Physical Review D, 26(6):1453–1478, 1982

    work page 1982

  39. [39]

    Curtis, M.R

    D.C. Curtis, M.R. Pennington, and D.A. Walsh. On the gauge dependence of dynamical fermion masses. Physics Letters B, 249(3):528–530, 1990

    work page 1990

  40. [40]

    Bashir, A

    A. Bashir, A. Kızılers¨ u, and M. R. Pennington. Gauge technique in qed. Physical Review D, 57(2):1242–1249, 1998

    work page 1998

  41. [41]

    V. M. B. Guzm´ an and A. Bashir. Nonperturbative structure of the fermion-boson vertex in qed. Physical Review D , 107(7):073008, 2023

    work page 2023

  42. [42]

    Light-cone current algebra and the introduction of color charge, 1971

    Harald Fritzsch and Murray Gell-Mann. Light-cone current algebra and the introduction of color charge, 1971

    work page 1971

  43. [43]

    D. J. Gross and F. Wilczek. Ultraviolet behavior of non-abelian gauge theories. Physical Review Letters , 30:1343–1346, 1973

    work page 1973

  44. [44]

    H. D. Politzer. Reliable perturbative results for strong interactions? Physical Review Letters, 30:1346–1349, 1973

    work page 1973

  45. [45]

    Perturbation theory without gauge fixing

    Giorgio Parisi and Yong-Shi Wu. Perturbation theory without gauge fixing. Science Sinica, 24:483–496, 1981

    work page 1981

  46. [46]

    Physics-informed neural networks viewpoint for solving the Dyson-Schwinger equations of quantum electrodynamics

    Rodrigo Carmo Terin. Physics-informed neural networks viewpoint for solving the Dyson-Schwinger equations of quantum electrodynamics. SciPost Phys. Core , 8:054, 2025

    work page 2025

  47. [47]

    P. C. Martin, E. D. Siggia, and H. A. Rose. Statistical dynamics of classical systems. Physical Review A , 8(1):423–437, 1973

    work page 1973

  48. [48]

    H.-K. Janssen. On a lagrangian for classical field dynamics and renormalization group calculations of dynamical critical properties. Zeitschrift f¨ ur Physik B - Condensed Matter, 23:377–380, 1976

    work page 1976

  49. [49]

    De Dominicis

    C. De Dominicis. A lagrangian version of halperin-hohenberg-ma models for the dynamics of critical phenomena. Lettere al Nuovo Cimento (1971-1985) , 12:567–574, 1975

    work page 1971

  50. [50]

    On the theory of brownian motion

    Paul Langevin. On the theory of brownian motion. CR Acad. Sci. (Paris) , 146:530–533, 1908. Reprinted in Am. J. Phys. 65 (1997) 1079

    work page 1908

  51. [51]

    Ba, and Nima Dehmamy

    Manos Theodosis, Demba E. Ba, and Nima Dehmamy. Constructing gauge-invariant neural networks for scientific appli- cations. ICML AI4Science Workshop , 2024

    work page 2024

  52. [52]

    Strogatz

    Steven H. Strogatz. Nonlinear Dynamics and Chaos: With Applications to Physics, Biology, Chemistry, and Engineering . 18 Westview Press, 2nd edition, 2014

    work page 2014

  53. [53]

    Christopher G. Langton. Computation at the edge of chaos: Phase transitions and emergent computation. Physica D: Nonlinear Phenomena, 42(1-3):12–37, 1990

    work page 1990

  54. [54]

    Real-time computation at the edge of chaos in recurrent neural networks

    Nils Bertschinger and Thomas Natschl¨ ager. Real-time computation at the edge of chaos in recurrent neural networks. Neural Computation, 16(7):1413–1436, 2004

    work page 2004

  55. [55]

    Buonomano

    Rodrigo Laje and Dean V. Buonomano. Robust timing and motor patterns by taming chaos in recurrent neural networks. Nature Neuroscience, 16(7):925–933, 2013

    work page 2013

  56. [56]

    L. D. Faddeev and V. N. Popov. Feynman diagrams for the yang–mills field. Physics Letters B , 25:29–30, 1967

    work page 1967

  57. [57]

    Hinton and Ruslan R

    Geoffrey E. Hinton and Ruslan R. Salakhutdinov. Reducing the dimensionality of data with neural networks. Science, 313(5786):504–507, 2006

    work page 2006

  58. [58]

    Stochastic integral.Proceedings of the Imperial Academy, 20:519–524, 1944

    Kiyosi Itˆ o. Stochastic integral.Proceedings of the Imperial Academy, 20:519–524, 1944

    work page 1944

  59. [59]

    American Mathematical Society, 1951

    Kiyosi Itˆ o.On stochastic differential equations , volume 4 of Memoirs of the American Mathematical Society . American Mathematical Society, 1951

    work page 1951

  60. [60]

    Kawai, D

    H. Kawai, D. C. Lewellen, and S.-H. H. Tye. A relation between tree amplitudes of closed and open strings. Nucl. Phys. B, 269:1–23, 1986

    work page 1986

  61. [61]

    Z. Bern, J. J. Carrasco, and H. Johansson. New relations for gauge-theory amplitudes. Phys. Rev. D , 78:085011, 2008

    work page 2008

  62. [62]

    Z. Bern, J. J. Carrasco, and H. Johansson. Perturbative quantum gravity as a double copy of gauge theory. Phys. Rev. Lett., 105:061602, 2010

    work page 2010

  63. [63]

    Z. Bern, J. J. Carrasco, M. Chiodaroli, H. Johansson, and R. Roiban. The duality between color and kinematics and its applications. J. Phys. A , 52:093001, 2019

    work page 2019

  64. [64]

    Sompolinsky, A

    H. Sompolinsky, A. Crisanti, and H. J. Sommers. Chaos in random neural networks. Phys. Rev. Lett., 61(3):259–262, 1988

    work page 1988

  65. [65]

    Exponential expressivity in deep neural networks through transient chaos

    Ben Poole, Subhaneil Lahiri, Maithra Raghu, Jascha Sohl-Dickstein, and Surya Ganguli. Exponential expressivity in deep neural networks through transient chaos. In Daniel D. Lee, Masashi Sugiyama, Ulrike V. Luxburg, Isabelle Guyon, and Roman Garnett, editors, Advances in Neural Information Processing Systems 29: Annual Conference on Neural Information Proc...

    work page 2016

  66. [66]

    Di Luo, Shunyue Yuan, James Stokes, and Bryan K. Clark. Gauge Equivariant Neural Networks for 2+1D U(1) Gauge Theory Simulations in Hamiltonian Formulation. 11 2022

    work page 2022

  67. [67]

    A. L. Hodgkin and A. F. Huxley. A quantitative description of membrane current and its application to conduction and excitation in nerve. The Journal of Physiology , 117(4):500–544, 1952

    work page 1952

  68. [68]

    McCulloch and Walter Pitts

    Warren S. McCulloch and Walter Pitts. A logical calculus of the ideas immanent in nervous activity. The Bulletin of Mathematical Biophysics, 5(4):115–133, 1943

    work page 1943

  69. [69]

    Donald O. Hebb. The Organization of Behavior: A Neuropsychological Theory . Wiley, 1949

    work page 1949

  70. [70]

    Churchland and Terrence J

    Patricia S. Churchland and Terrence J. Sejnowski. The Computational Brain . MIT Press, 1992

    work page 1992

  71. [71]

    Spikes: Exploring the Neural Code

    Fred Rieke, David Warland, Rob de Ruyter van Steveninck, and William Bialek. Spikes: Exploring the Neural Code . MIT Press, 1999

    work page 1999

  72. [72]

    Mountcastle

    Vernon B. Mountcastle. An organizing principle for cerebral function: the unit model and the distributed system. The Mindful Brain, pages 7–50, 1978

    work page 1978

  73. [73]

    Felleman and David C

    Daniel J. Felleman and David C. Van Essen. Distributed hierarchical processing in the primate cerebral cortex. Cerebral Cortex, 1(1):1–47, 1991

    work page 1991

  74. [74]

    Giulio Tononi, Olaf Sporns, and Gerald M. Edelman. A measure for brain complexity: relating functional segregation and integration in the nervous system. Proceedings of the National Academy of Sciences , 91(11):5033–5037, 1994

    work page 1994

  75. [75]

    A mechanism for cognitive dynamics: neuronal communication through neuronal coherence

    Pascal Fries. A mechanism for cognitive dynamics: neuronal communication through neuronal coherence. Trends in Cognitive Sciences, 9(10):474–480, 2005

    work page 2005

  76. [76]

    Vision: A Computational Investigation into the Human Representation and Processing of Visual Information

    David Marr. Vision: A Computational Investigation into the Human Representation and Processing of Visual Information . W. H. Freeman, 1982

    work page 1982

  77. [77]

    Hopfield

    John J. Hopfield. Neural networks and physical systems with emergent collective computational abilities. Proceedings of the National Academy of Sciences , 79(8):2554–2558, 1982

    work page 1982

  78. [78]

    O’Reilly and Yuko Munakata

    Randall C. O’Reilly and Yuko Munakata. Computational Explorations in Cognitive Neuroscience . MIT Press, 2000

    work page 2000

  79. [79]

    The free-energy principle: a unified brain theory? Nature Reviews Neuroscience, 11(2):127–138, 2010

    Karl Friston. The free-energy principle: a unified brain theory? Nature Reviews Neuroscience, 11(2):127–138, 2010

    work page 2010

  80. [80]

    Computational complexity and fundamental limitations to fermionic quantum monte carlo simulations

    Matthias Troyer and Uwe-Jens Wiese. Computational complexity and fundamental limitations to fermionic quantum monte carlo simulations. Physical Review Letters, 94(17):170201, 2005

    work page 2005