pith. sign in

arxiv: 2606.07247 · v2 · pith:N5P752ZZnew · submitted 2026-06-05 · ❄️ cond-mat.dis-nn · cond-mat.stat-mech· stat.ML

Theory of learning of high-dimensional controlled non-linear dynamical systems (I): models and methods

Pierfrancesco Urbani This is my paper

Pith reviewed 2026-06-27 20:22 UTC · model grok-4.3

classification ❄️ cond-mat.dis-nn cond-mat.stat-mechstat.ML
keywords neural ODEsdynamical mean field theoryhigh-dimensional limitonline stochastic gradient descentlearning curvestraining dynamicscontrolled dynamical systemsgeneralization
0
0 comments X

The pith

A class of controlled non-linear dynamical systems allows exact solution of neural ODE training dynamics in the high-dimensional limit via dynamical mean field theory.

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

The paper introduces a theoretically grounded class of controlled non-linear dynamical systems to model neural ODEs trained by online stochastic gradient descent. It applies dynamical mean field theory to close the equations for the dual inference and training dynamics. This yields explicit learning curves in the high-dimensional limit. The framework covers settings such as ResNets, autoregressive models, generative models, and recurrent networks in neuroscience. A sympathetic reader would care because it turns the training of continuous-time neural networks into a solvable problem rather than a black-box optimization task.

Core claim

We introduce a theoretically grounded class of models for studying neural ODEs trained via online stochastic gradient descent. We solve the training dynamics of these models via dynamical mean field theory and derive learning curves in the high-dimensional limit.

What carries the argument

Dynamical mean field theory closure on the coupled inference and training dynamics of the controlled non-linear systems.

Load-bearing premise

The introduced class of controlled non-linear dynamical systems is representative enough that the mean-field equations capture the essential training behavior of the wider family of neural ODEs.

What would settle it

A direct numerical simulation of a neural ODE outside the controlled class whose measured training trajectory deviates from the predicted learning curve in the high-dimensional regime.

Figures

Figures reproduced from arXiv: 2606.07247 by Pierfrancesco Urbani.

Figure 1
Figure 1. Figure 1: The pipeline of training dynamics. condition. The algorithm requires storing the initial and final states of the Teacher process, x(0, α) and x(Ts, α), which constitute the α-th training point, as well as the entire trajectory of the Student process, y(t, α), for t ∈ [0, Ts]. Following this, the dynamics of the adjoint fields π are computed backward in time. Once the adjoint trajectory is obtained, the gra… view at source ↗
Figure 2
Figure 2. Figure 2: Comparison between DMFT and numerical simulations. We plot tree observables [PITH_FULL_IMAGE:figures/full_fig_p023_2.png] view at source ↗
read the original abstract

Neural ordinary differential equations (neural ODEs) have rapidly gained prominence as a powerful and unifying framework for conceptualizing artificial neural networks, elegantly connecting the continuous-time modeling of dynamical systems with the discrete, data-driven paradigm of modern deep learning. Beyond their practical advantages they offer fresh theoretical insights into the training and generalization properties of neural networks. The distinctive feature of this framework is its dual dynamical nature: inference dynamics, which govern the ODE evolution during forward computation, and training dynamics, which control the optimization of model parameters. This makes neural ODEs a particularly well-suited theoretical framework for studying a large variety of settings such as multi-layer neural networks (ResNets for example), autoregressive models (with next-token generation dynamics), generative models, and recurrent neural networks in theoretical neuroscience. In this work, we introduce a theoretically grounded class of models for studying neural ODEs trained via online stochastic gradient descent. We solve the training dynamics of these models via dynamical mean field theory and derive learning curves in the high-dimensional limit.

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 / 0 minor

Summary. The paper introduces a class of controlled non-linear dynamical systems as a theoretically grounded model for neural ODEs (including ResNets, autoregressive models, RNNs, and generative models) trained by online SGD. It claims to solve the training dynamics exactly via dynamical mean-field theory (DMFT) and to derive closed learning curves in the high-dimensional limit.

Significance. If the DMFT closure is exact for the proposed class and the class is representative of standard neural-ODE architectures, the work would supply the first parameter-free, high-dimensional learning curves for a broad family of continuous-depth models, a substantial advance over existing heuristic or simulation-based analyses.

major comments (2)
  1. [Abstract and §1] Abstract and §1: the central claim that the introduced controlled non-linear class is sufficiently representative for the DMFT closure to capture essential forward and training dynamics of the broader family of neural ODEs (ResNets, autoregressive models, RNNs) is asserted without an explicit mapping, specialization check, or numerical comparison showing that the closure survives when the model is reduced to standard neural-ODE forms. This representativeness step is load-bearing for the applicability statements.
  2. [Abstract] Abstract: the assertion that DMFT 'yields closed learning curves' is stated without any displayed order-parameter equations, closure ansatz, or finite-N validation; the absence of these elements in the provided text prevents assessment of whether the claimed closure is actually achieved or merely conjectured.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading and constructive feedback. We respond to each major comment below.

read point-by-point responses

  1. Referee: [Abstract and §1] Abstract and §1: the central claim that the introduced controlled non-linear class is sufficiently representative for the DMFT closure to capture essential forward and training dynamics of the broader family of neural ODEs (ResNets, autoregressive models, RNNs) is asserted without an explicit mapping, specialization check, or numerical comparison showing that the closure survives when the model is reduced to standard neural-ODE forms. This representativeness step is load-bearing for the applicability statements.

    Authors: We agree that explicit mappings would strengthen the claims. In the revised manuscript we will add a subsection in §1 providing explicit reductions of the controlled non-linear class to ResNets, RNNs and autoregressive models, together with the corresponding specializations of the DMFT equations and numerical checks confirming that the closure is preserved under these reductions. revision: yes

  2. Referee: [Abstract] Abstract: the assertion that DMFT 'yields closed learning curves' is stated without any displayed order-parameter equations, closure ansatz, or finite-N validation; the absence of these elements in the provided text prevents assessment of whether the claimed closure is actually achieved or merely conjectured.

    Authors: The order-parameter equations, closure ansatz and finite-N validations appear in Sections 3–5. The abstract summarizes the result at a high level, which is standard. To improve clarity we will insert a concise outline of the key DMFT equations and ansatz into the introduction of the revised version. revision: partial

Circularity Check

0 steps flagged

No circularity: DMFT applied to newly introduced model class

full rationale

The paper introduces a class of controlled non-linear dynamical systems and applies dynamical mean-field theory to derive its training dynamics and learning curves in the high-d limit. No quoted step reduces a result to a fitted parameter, self-definition, or self-citation chain; the derivation is presented as a standard closure on the defined model without evidence that predictions are tautological with inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the applicability of dynamical mean-field theory to the newly defined models; no free parameters or invented entities are mentioned in the abstract.

axioms (1)
  • domain assumption Dynamical mean field theory provides an exact closure for the training dynamics of the introduced high-dimensional models.
    The abstract states that the training dynamics are solved via DMFT, which presupposes that the mean-field approximation becomes exact in the high-dimensional limit.

pith-pipeline@v0.9.1-grok · 5714 in / 1230 out tokens · 21578 ms · 2026-06-27T20:22:46.574908+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

57 extracted references · 4 linked inside Pith

  1. [1]

    Learning representations by back-propagating errors.Nature, 323(6088):533–536, 1986

    David E Rumelhart, Geoffrey E Hinton, and Ronald J Williams. Learning representations by back-propagating errors.Nature, 323(6088):533–536, 1986. 2

    1986

  2. [2]

    Multilayer feedforward networks are universal approximators.Neural networks, 2(5):359–366, 1989

    Kurt Hornik, Maxwell B Stinchcombe, and Halbert White. Multilayer feedforward networks are universal approximators.Neural networks, 2(5):359–366, 1989. 2 24

    1989

  3. [3]

    MIT Press, Cambridge, MA, USA, 2021

    Yoshua Bengio, Ian Goodfellow, and Aaron Courville.Deep Learning. MIT Press, Cambridge, MA, USA, 2021. 2

    2021

  4. [4]

    Deep convolutional neural networks for image classification: A comprehensive review.Neural Computation, 29(9):2352–2449, 2017

    Wang Rawat and Zan Wang. Deep convolutional neural networks for image classification: A comprehensive review.Neural Computation, 29(9):2352–2449, 2017. 2

    2017

  5. [5]

    Springer science & business media,

    Vladimir Vapnik.The nature of statistical learning theory. Springer science & business media,

  6. [6]

    Understand- ing deep learning requires rethinking generalization.arXiv preprint arXiv:1611.03530, 2016

    Chiyuan Zhang, Samy Bengio, Moritz Hardt, Benjamin Recht, and Oriol Vinyals. Understand- ing deep learning requires rethinking generalization.arXiv preprint arXiv:1611.03530, 2016. 3

    Pith/arXiv arXiv 2016

  7. [7]

    Under- standing deep learning (still) requires rethinking generalization.Communications of the ACM, 64(3):107–115, 2021

    Chiyuan Zhang, Samy Bengio, Moritz Hardt, Benjamin Recht, and Oriol Vinyals. Under- standing deep learning (still) requires rethinking generalization.Communications of the ACM, 64(3):107–115, 2021. 3

    2021

  8. [8]

    Neural tangent kernel: Convergence and generalization in neural networks.arXiv preprint arXiv:1806.07572, 2018

    Arthur Jacot, Fran¸ cois Gabriel, and Cl´ ement Hongler. Neural tangent kernel: Convergence and generalization in neural networks.arXiv preprint arXiv:1806.07572, 2018. 3

    arXiv 2018

  9. [9]

    On the global convergence of gradient descent for over- parameterized models with smooth activations.arXiv preprint arXiv:1806.02629, 2018

    Lenaic Chizat and Francis Bach. On the global convergence of gradient descent for over- parameterized models with smooth activations.arXiv preprint arXiv:1806.02629, 2018. 3

    arXiv 2018

  10. [10]

    Six lectures on linearized models.arXiv preprint, 2023

    Andrea Montanari. Six lectures on linearized models.arXiv preprint, 2023. 3

    2023

  11. [11]

    A mean field view of the landscape of two-layer neural networks.Proceedings of the National Academy of Sciences, 115(33):E7665– E7671, 2018

    Song Mei, Andrea Montanari, and Phan-Minh Nguyen. A mean field view of the landscape of two-layer neural networks.Proceedings of the National Academy of Sciences, 115(33):E7665– E7671, 2018. 3

    2018

  12. [12]

    Trainability and accuracy of artificial neural net- works: An interacting particle system approach.Communications on Pure and Applied Math- ematics, 75(9):1889–1935, 2022

    Grant Rotskoff and Eric Vanden-Eijnden. Trainability and accuracy of artificial neural net- works: An interacting particle system approach.Communications on Pure and Applied Math- ematics, 75(9):1889–1935, 2022. 3

    1935

  13. [13]

    Dynamical decoupling of generalization and overfitting in large two-layer networks.arXiv preprint arXiv:2502.21269, 2025

    Andrea Montanari and Pierfrancesco Urbani. Dynamical decoupling of generalization and overfitting in large two-layer networks.arXiv preprint arXiv:2502.21269, 2025. 3, 4, 11, 12

    arXiv 2025

  14. [14]

    Neural ordinary differential equations

    Ricky TQ Chen, Yulia Rubanova, Jesse Bettencourt, and David Duvenaud. Neural ordinary differential equations. InAdvances in Neural Information Processing Systems, pages 6571– 6583, 2018. 3

    2018

  15. [15]

    On neural differential equations.arXiv preprint arXiv:2202.02435, 2022

    Patrick Kidger. On neural differential equations.arXiv preprint arXiv:2202.02435, 2022. 3

    arXiv 2022

  16. [16]

    Attention is all you need.Advances in Neural Information Processing Systems, 2017

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in Neural Information Processing Systems, 2017. 3

    2017

  17. [17]

    Language models are unsupervised multitask learners.OpenAI Blog, 2019

    Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. Language models are unsupervised multitask learners.OpenAI Blog, 2019. 3

    2019

  18. [18]

    Efficiently modeling long sequences with struc- tured state spaces.arXiv preprint arXiv:2111.00396, 2021

    Albert Gu, Karan Goel, and Christopher R´ e. Efficiently modeling long sequences with struc- tured state spaces.arXiv preprint arXiv:2111.00396, 2021. 3 25

    Pith/arXiv arXiv 2021

  19. [19]

    Hippo: Recurrent memory with optimal polynomial projections.Neural Information Processing Systems, 2020

    Albert Gu, Tri Dao, Stephen Gu, Daniel Dohan, Emily Chen, Rewon Child, and Christopher R´ e. Hippo: Recurrent memory with optimal polynomial projections.Neural Information Processing Systems, 2020. 3

    2020

  20. [20]

    FFJORD: Free-form continuous dynamics for reversible generative models.arXiv preprint arXiv:1810.01367, 2018

    Will Grathwohl, Ricky TQ Chen, Jesse Betancourt, Jascha Sohl-Dickstein, and David Du- venaud. FFJORD: Free-form continuous dynamics for reversible generative models.arXiv preprint arXiv:1810.01367, 2018. 3

    Pith/arXiv arXiv 2018

  21. [21]

    Harnessing Nonlinearity: Predicting chaotic systems and saving energy with reservoir computing.Science, 304(5667):78–80, 2004

    Herbert Jaeger and Harro Haas. Harnessing Nonlinearity: Predicting chaotic systems and saving energy with reservoir computing.Science, 304(5667):78–80, 2004. 3

    2004

  22. [22]

    Real-time computing without stable states: A new framework for neural computation based on perturbations.Neural Com- putation, 14(11):2531–2560, 2002

    Wolfgang Maass, Thomas Natschl¨ ager, and Henry Markram. Real-time computing without stable states: A new framework for neural computation based on perturbations.Neural Com- putation, 14(11):2531–2560, 2002. 3

    2002

  23. [23]

    Generating coherent patterns of activity from chaotic neural networks.Neuron, 63(4):544–557, 2009

    David Sussillo and Larry F Abbott. Generating coherent patterns of activity from chaotic neural networks.Neuron, 63(4):544–557, 2009. 3, 4

    2009

  24. [24]

    Three unfinished works on the optimal storage capacity of networks.Journal of Physics A: Mathematical and General, 22(12):1983–1994,

    Elizabeth Gardner and Bernard Derrida. Three unfinished works on the optimal storage capacity of networks.Journal of Physics A: Mathematical and General, 22(12):1983–1994,

    1983

  25. [25]

    World Scientific Publishing Company, 1987

    Marc M´ ezard, Giorgio Parisi, and Miguel Angel Virasoro.Spin glass theory and beyond: An Introduction to the Replica Method and Its Applications, volume 9. World Scientific Publishing Company, 1987. 4, 15, 16

    1987

  26. [26]

    Passed & spurious: Descent algorithms and local minima in spiked matrix-tensor models

    Stefano Sarao Mannelli, Florent Krzakala, Pierfrancesco Urbani, and Lenka Zdeborova. Passed & spurious: Descent algorithms and local minima in spiked matrix-tensor models. Ininterna- tional conference on machine learning, pages 4333–4342. PMLR, 2019. 4

    2019

  27. [27]

    Marvels and pitfalls of the langevin algorithm in noisy high- dimensional inference.Physical Review X, 10(1):011057, 2020

    Stefano Sarao Mannelli, Giulio Biroli, Chiara Cammarota, Florent Krzakala, Pierfrancesco Urbani, and Lenka Zdeborov´ a. Marvels and pitfalls of the langevin algorithm in noisy high- dimensional inference.Physical Review X, 10(1):011057, 2020. 4

    2020

  28. [28]

    Analytical study of momentum-based accel- eration methods in paradigmatic high-dimensional non-convex problems.Advances in Neural Information Processing Systems, 34:187–199, 2021

    Stefano Sarao Mannelli and Pierfrancesco Urbani. Analytical study of momentum-based accel- eration methods in paradigmatic high-dimensional non-convex problems.Advances in Neural Information Processing Systems, 34:187–199, 2021. 4

    2021

  29. [29]

    Francesca Mignacco, Pierfrancesco Urbani, and Lenka Zdeborov´ a. Stochasticity helps to nav- igate rough landscapes: comparing gradient-descent-based algorithms in the phase retrieval problem.Machine Learning: Science and Technology, 2(3):035029, 2021. 4

    2021

  30. [30]

    Stochastic gradient descent outperforms gradi- ent descent in recovering a high-dimensional signal in a glassy energy landscape.arXiv preprint arXiv:2309.04788, 2023

    Persia Jana Kamali and Pierfrancesco Urbani. Stochastic gradient descent outperforms gradi- ent descent in recovering a high-dimensional signal in a glassy energy landscape.arXiv preprint arXiv:2309.04788, 2023. 4, 11, 12

    arXiv 2023

  31. [31]

    Out-of- equilibrium dynamical mean-field equations for the perceptron model.Journal of Physics A: Mathematical and Theoretical, 51(8):085002, 2018

    Elisabeth Agoritsas, Giulio Biroli, Pierfrancesco Urbani, and Francesco Zamponi. Out-of- equilibrium dynamical mean-field equations for the perceptron model.Journal of Physics A: Mathematical and Theoretical, 51(8):085002, 2018. 4, 12

    2018

  32. [32]

    Dynam- ical mean-field theory for stochastic gradient descent in gaussian mixture classification.Ad- vances in Neural Information Processing Systems, 33:9540–9550, 2020

    Francesca Mignacco, Florent Krzakala, Pierfrancesco Urbani, and Lenka Zdeborov´ a. Dynam- ical mean-field theory for stochastic gradient descent in gaussian mixture classification.Ad- vances in Neural Information Processing Systems, 33:9540–9550, 2020. 4, 12 26

    2020

  33. [33]

    The estimation error of general first order methods

    Michael Celentano, Andrea Montanari, and Yuchen Wu. The estimation error of general first order methods. InConference on Learning Theory, pages 1078–1141. PMLR, 2020. 4

    2020

  34. [34]

    The effective noise of stochastic gradient de- scent.Journal of Statistical Mechanics: Theory and Experiment, 2022(8):083405, 2022

    Francesca Mignacco and Pierfrancesco Urbani. The effective noise of stochastic gradient de- scent.Journal of Statistical Mechanics: Theory and Experiment, 2022(8):083405, 2022. 4, 12

    2022

  35. [35]

    A dynamical model of neural scaling laws.arXiv preprint arXiv:2402.01092, 2024

    Blake Bordelon, Alexander Atanasov, and Cengiz Pehlevan. A dynamical model of neural scaling laws.arXiv preprint arXiv:2402.01092, 2024. 4

    arXiv 2024

  36. [36]

    Dynamical mean-field analysis of adaptive langevin diffusions: Propagation-of-chaos and convergence of the linear response.arXiv preprint arXiv:2504.15556, 2025

    Zhou Fan, Justin Ko, Bruno Loureiro, Yue M Lu, and Yandi Shen. Dynamical mean-field analysis of adaptive langevin diffusions: Propagation-of-chaos and convergence of the linear response.arXiv preprint arXiv:2504.15556, 2025. 4

    arXiv 2025

  37. [37]

    Statistical physics of learning in high- dimensional chaotic systems.Journal of Statistical Mechanics: Theory and Experiment, 2023(11):113301, 2023

    Samantha J Fournier and Pierfrancesco Urbani. Statistical physics of learning in high- dimensional chaotic systems.Journal of Statistical Mechanics: Theory and Experiment, 2023(11):113301, 2023. 4, 5, 6

    2023

  38. [38]

    Structure, disorder, and dynamics in task-trained recurrent neural circuits.bioRxiv, pages 2026–03,

    David G Clark, Blake Bordelon, Jacob A Zavatone-Veth, and Cengiz Pehlevan. Structure, disorder, and dynamics in task-trained recurrent neural circuits.bioRxiv, pages 2026–03,

    2026

  39. [39]

    To appear

    Samantha Fournier and Pierfrancesco Urbani. To appear. 2026. 4

    2026

  40. [40]

    Disordered high-dimensional optimal control.Journal of Physics A: Mathematical and Theoretical, 54(32):324001, 2021

    Pierfrancesco Urbani. Disordered high-dimensional optimal control.Journal of Physics A: Mathematical and Theoretical, 54(32):324001, 2021. 4

    2021

  41. [41]

    Unpublished

    Yuri Lombardo and Pierfrancesco Urbani. Unpublished. see: Yuri, lombardo optimization of the gradient descent dynamics in simple mean field spin glasses. 2021. 4

    2021

  42. [42]

    Optimal protocols for continual learning via statistical physics and control theory.Journal of Statistical Mechanics: Theory and Experiment, 2025(8):084004, 2025

    Francesco Mori, Stefano Sarao Mannelli, and Francesca Mignacco. Optimal protocols for continual learning via statistical physics and control theory.Journal of Statistical Mechanics: Theory and Experiment, 2025(8):084004, 2025. 4

    2025

  43. [43]

    Macroscopic fluctuation theory.Reviews of Modern Physics, 87(2):593–636, 2015

    Lorenzo Bertini, Alberto De Sole, Davide Gabrielli, Giovanni Jona-Lasinio, and Claudio Landim. Macroscopic fluctuation theory.Reviews of Modern Physics, 87(2):593–636, 2015. 5

    2015

  44. [44]

    Quantum optimal control theory.Journal of Physics B: Atomic, Molecular and Optical Physics, 40(18):R175–R211, 2007

    J Werschnik and EKU Gross. Quantum optimal control theory.Journal of Physics B: Atomic, Molecular and Optical Physics, 40(18):R175–R211, 2007. 5

    2007

  45. [45]

    Non- reciprocal interactions and high-dimensional chaos: comparing dynamics and statistics of equi- libria in a solvable model

    Samantha J Fournier, Alessandro Pacco, Valentina Ros, and Pierfrancesco Urbani. Non- reciprocal interactions and high-dimensional chaos: comparing dynamics and statistics of equi- libria in a solvable model. 2025. 5, 6

    2025

  46. [46]

    High-dimensional dynamical systems: co- existence of attractors, phase transitions, maximal lyapunov exponent and response to periodic drive.arXiv preprint arXiv:2511.09679, 2025

    Samantha J Fournier and Pierfrancesco Urbani. High-dimensional dynamical systems: co- existence of attractors, phase transitions, maximal lyapunov exponent and response to periodic drive.arXiv preprint arXiv:2511.09679, 2025. 5, 6

    arXiv 2025

  47. [47]

    Chaos in high-dimensional dynamical systems with tunable non-reciprocity.arXiv preprint arXiv:2601.04702, 2026

    Samantha Fournier and Pierfrancesco Urbani. Chaos in high-dimensional dynamical systems with tunable non-reciprocity.arXiv preprint arXiv:2601.04702, 2026. 6

    Pith/arXiv arXiv 2026

  48. [48]

    Backpropagation and the brain.Nature Reviews Neuroscience, 21(6):335–346, 2020

    Timothy P Lillicrap, Adam Santoro, Luke Marris, Colin J Akerman, and Geoffrey Hinton. Backpropagation and the brain.Nature Reviews Neuroscience, 21(6):335–346, 2020. 8 27

    2020

  49. [49]

    A mean-field optimal control formulation of deep learning.arXiv preprint arXiv:1807.01083, 2018

    E Weinan, Jiequn Han, and Qianxiao Li. A mean-field optimal control formulation of deep learning.arXiv preprint arXiv:1807.01083, 2018. 10

    arXiv 2018

  50. [50]

    Dynamical mean field theory for models of confluent tissues and beyond.SciPost Physics, 15(5):219, 2023

    Persia Jana Kamali and Pierfrancesco Urbani. Dynamical mean field theory for models of confluent tissues and beyond.SciPost Physics, 15(5):219, 2023. 11

    2023

  51. [51]

    Emergence and scaling laws in sgd learning of shallow neural networks.Advances in Neural Information Processing Systems, 38:38227–38309, 2026

    Yunwei Ren, Eshaan Nichani, Denny Wu, and Jason Lee. Emergence and scaling laws in sgd learning of shallow neural networks.Advances in Neural Information Processing Systems, 38:38227–38309, 2026. 11

    2026

  52. [52]

    Statistical dynamics of classical systems

    Paul Cecil Martin, Eric D Siggia, and Harvey A Rose. Statistical dynamics of classical systems. Physical Review A, 8(1):423, 1973. 12

    1973

  53. [53]

    Hans-Karl Janssen. On a lagrangean for classical field dynamics and renormalization group calculations of dynamical critical properties.Zeitschrift f¨ ur Physik B Condensed Matter, 23(4):377–380, 1976. 12

    1976

  54. [54]

    Techniques de renormalisation de la th´ eorie des champs et dynamique des ph´ enomenes critiques.J

    C De Dominicis. Techniques de renormalisation de la th´ eorie des champs et dynamique des ph´ enomenes critiques.J. Phys. Colloques, 37, 1976. 12

    1976

  55. [55]

    Dynamics as a substitute for replicas in systems with quenched random impurities.Physical Review B, 18(9):4913, 1978

    C De Dominicis. Dynamics as a substitute for replicas in systems with quenched random impurities.Physical Review B, 18(9):4913, 1978. 12, 13

    1978

  56. [56]

    Oxford university press, 2021

    Jean Zinn-Justin.Quantum field theory and critical phenomena, volume 171. Oxford university press, 2021. 12

    2021

  57. [57]

    Diffusion with stochastic resetting.Physical review letters, 106(16):160601, 2011

    Martin R Evans and Satya N Majumdar. Diffusion with stochastic resetting.Physical review letters, 106(16):160601, 2011. 19 28

    2011