Pith. sign in

REVIEW 2 major objections 4 minor 67 references

The Kinetic Limit of Balanced Neural Networks

T0 review · 2 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Balanced excitation–inhibition networks have a rigorous kinetic limit.

desk verdict Genuine progress on kinetic limits of balanced networks, but the proof has a false independence claim in a central lemma and needs revision before it can be trusted. read the letter →

arxiv 2505.18481 v1 pith:FNM7CRRO submitted 2025-05-24 math.PR math.DS

James MacLaurin , Pedro Vilanova This is my paper
classification math.PRmath.DS MSC 60K3560H1092B20
keywords balancedneuralnetworkskineticlimitmean-fieldFokker-Planckequationmanifoldn^{-1/2}interactionsfieldequationsmultiplicativenoise
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper aims to prove that the balanced state of cortical networks, where excitation and inhibition nearly cancel, survives as a genuine large-population limit rather than only as a time-averaged ansatz. For a network of $n$ excitatory and $n$ inhibitory neurons with interactions of order $n^{-1/2}$, nonlinear intrinsic dynamics, multiplicative noise, and spatially dependent finite-rank connectivity, it shows that the empirical distribution of fluctuations and the mean activities converge almost surely to unique limiting quantities. The limit consists of coupled stochastic differential equations for single-neuron fluctuations together with the implicit balance condition $G(\bar{v}, \mu_t) = 0$, which fixes the mean activity, while the fluctuation density evolves by a Fokker-Planck equation. This matters because it converts the balanced-network explanation of cortical variability into a theorem about an autonomous dynamical system that is valid out of equilibrium, and in a linear special case it reduces to a spatially distributed neural field equation.

What carries the argument

The load-bearing object is the Balanced Manifold $\mathcal{B}$, the set of pairs $(v,\mu)$ for which the balance map $G(v,\mu)$ vanishes, every eigenvalue of the Jacobian $J(v,\mu)$ has strictly negative real part, and the spatial marginal of $\mu$ equals the limiting neuron-position law $\kappa$. The map $G$ integrates the difference between excitatory and inhibitory inputs against each basis function $h_a$; its vanishing is what makes the $n^{-1/2}$ interaction terms cancel on average. The argument decomposes each neuron's state into a mean-field component in the span of the basis functions plus a fluctuation $y$, tracks the empirical fluctuation measure $\hat{\mu}_t^n$, and uses the spectral gap of $J$ to show that the projected mean $v(t)$ is pulled toward $\bar{v}(t)$ faster than fluctuations can push it away, as quantified in Lemma 6.1 and Corollary 6.2. In the limit, fluctuations obey the coupled SDEs (3.2)--(3.3) with density solving the Fokker-Planck equation, and the mean obeys the implicit ODE $d\bar{v}/dt = -J^{-1}H$ from Corollary 3.3.

What would settle it

Choose a finite-rank connectivity kernel and parameters for which the Jacobian $J$ of the balance map develops an eigenvalue with positive real part before a time $T$, and run the particle system (2.5)--(2.6) for increasing $n$. If the empirical fluctuation measure and mean activities still converge to a balanced limit beyond that time, the spectral-gap hypothesis is not necessary; if instead fluctuations grow or oscillations appear with frequency increasing in $n$, the theorem's premise is load-bearing.

Watch

Extended reading notes

Core claim

The paper's central claim is that a network of excitatory and inhibitory neurons with interaction strengths scaled as $n^{-1/2}$, nonlinear intrinsic dynamics, multiplicative noise, and finite-rank spatial connectivity has a well-defined kinetic limit. Writing each neuron's activity as a mean-field projection onto finitely many spatial basis functions plus a fluctuation, the paper defines an empirical fluctuation measure $\hat{\mu}_t^n$ and mean coefficients $v(t)$. Theorem 3.2 states that for every time $T$ before the maximal balanced time $\eta$, almost surely $\hat{\mu}_t^n$ converges in Wasserstein distance to a unique measure $\mu_t$ and $v(t)$ converges to $\bar{v}(t)$, where $\mu_t$ is the law of the coupled SDEs (3.2)--(3.3) and the balance condition $G(\bar{v},\mu_t)=0$ holds for all $t<\eta$. Corollary 3.3 rewrites the mean dynamics as $d\bar{v}/dt = -J^{-1}H$, with $J$ the Jacobian of the balance map, and the density of $\mu_t$ solves the Fokker-Planck equation (3.11). Balance is therefore not assumed a priori: it is the attracting manifold selected by the limit, and the limiting equations are autonomous.

Load-bearing premise

The argument rests on the balance condition being linearly stable: small shifts in mean excitation or inhibition must decay exponentially, and the initial condition must lie exactly on the balanced manifold; without that spectral gap, the $O(n^{-1/2})$ interactions are no longer controlled.

Editorial extensions

If this is right

  • For every time before the maximal balanced time $\eta$, the empirical fluctuation measure converges almost surely in Wasserstein distance to the unique $\mu_t$, so finite-$n$ simulations can be compared directly with a closed system of equations.
  • The mean population activities are not free variables: they are slaved to the fluctuation law by $G(\bar{v},\mu_t)=0$, so asymptotic theories of balanced cortical variability must solve the coupled SDE-balance system rather than prescribe the means separately.
  • When the intrinsic dynamics is linear and the noise is additive, the limiting law is Gaussian and the system reduces to ODEs for local covariances together with a neural-field equation for the means, making pattern-formation questions tractable.
  • The maximal balanced time $\eta$ marks the limit of validity of the theorem: before $\eta$ the spectral gap persists and convergence holds, while at or after $\eta$ the theorem gives no control, consistent with the oscillatory divergence observed in the spatial simulations.

Reading between the lines

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

  • The proof's rates suggest a concrete finite-size scaling: deviations of the mean activity from its limit should be no larger than order $n^{-1/4}$ up to logarithmic factors whenever the balanced manifold is linearly stable; measuring that exponent in simulations would directly test the contraction mechanism.
  • The breakdown at $\eta$ looks like a dynamical phase transition in the mean-field equations: when $J$ loses negative definiteness, the implicit ODE for $\bar{v}$ can bifurcate, so balanced networks may switch between asynchronous and oscillatory regimes as connectivity parameters vary.
  • The finite-rank assumption restricts the mean-field component to finitely many spatial modes; replacing it with a general connectivity kernel would presumably force an infinite-dimensional balance map, and the same spectral-gap argument should still work if the associated operator has a uniform negative spectral bound.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 4 minor

Summary. The manuscript studies a system of 2n coupled stochastic differential equations modeling excitatory and inhibitory neurons, with interaction strengths scaled as n^{-1/2} and a spatial structure encoded by finite-rank kernels. The main result, Theorem 3.2, states that, for times below a maximal balanced time η, the empirical fluctuation measure and the empirical mean activities converge almost surely to a unique kinetic limit: a Fokker-Planck equation for the fluctuation density coupled, through a balance condition G(v̄, μ)=0, to an autonomous equation for the mean activity v̄. The proof strategy decomposes each neuron's state into a projected mean component and fluctuations, then bounds the fluctuation growth via the spectral gap of the Jacobian of the balance map. The paper also derives a covariance ODE and a neural-field reduction under linear intrinsic dynamics and additive noise, and provides numerical experiments comparing particle simulations with the limiting ODEs in both a linear inhibition-stabilized network and a nonlinear-gain network.

Significance. If the proof gaps identified below are repaired, this would be a notable contribution: it is, to the authors' knowledge, the first rigorous kinetic limit for balanced neural networks with nonlinear intrinsic dynamics and multiplicative noise, and it clarifies the role of linear stability of the balanced manifold. The explicit autonomous evolution of the mean (Corollary 3.3) and the reduction to a Gaussian neural-field equation are valuable for applications. The paper is also honest about the limitations of the spectral assumption, and the numerical experiments in Sections 4.1 and 4.2 provide a useful sanity check. The main concern is that a central step in the proof of the theorem, Lemma 6.4, is not justified as written, and the bootstrap in the proof of Theorem 3.2 needs clarification; these issues are load-bearing for the main claim.

major comments (2)
  1. [6, Lemma 6.4 (Eqs. (6.14)–(6.19))] The proof of Lemma 6.4 asserts that the processes (\bar y^j_e,\bar y^j_i)_{j\in I_n} are independent C([0,T],R^2)-valued random variables and then invokes the Law of Large Numbers. This independence claim is false: the drifts in (6.14)–(6.15) contain the common empirical average n^{-1}\sum_{k\in I_n}\sum_{a,b} Q^{-1}_{ab} f_\alpha(\bar z^k_{\alpha,t}) h_b(x^k_n) h_a(x^j_n), which couples each \bar y^j to all other particles through the empirical measure. The Brownian motions are independent, but the processes themselves are not, so the one-line LLN argument for (6.19) is invalid. Since (6.19) is used in the triangle inequality (6.22) to prove Theorem 3.2, this is a genuine gap. The convergence is likely recoverable by a standard propagation-of-chaos estimate for the empirical measure of weakly interacting diffusions, but the proof as written is incomplete.
  2. [6, proof of Theorem 3.2 and Lemma 6.5] The proof of Theorem 3.2 uses two stopping times: ζ_n in (6.20), which appears in Lemma 6.5, and φ, the first time √q_t = ε_T. The text states that Corollary 6.2 and Lemma 6.5 imply the bounds (6.28)–(6.29) for all t ≤ φ, but it does not justify that the hypotheses of those results are satisfied simultaneously on the event used to conclude (6.30). In particular, Corollary 6.2 only controls √ℓ_t on the event that sup_{s≤t} d_W(\hatµ^n_s, µ_s) ≤ ε_T and sup_{s≤t} ℓ_s ≤ ε_T, while Lemma 6.5 only controls q_t up to ζ_n, which itself depends on ℓ. A bootstrap or inductive argument is needed to show that these thresholds are not crossed before time T; as written, the simultaneous use of (6.28)–(6.29) for all t ≤ φ is not rigorously justified.
minor comments (4)
  1. [6, Eqs. (6.5)–(6.6)] The drift terms in (6.5) and (6.6) write f(z^j_e,t) and f(z^j_i,t) without subscripts; these should be f_e and f_i respectively to match the notation in the model.
  2. [Introduction, paragraph on Erny et al.] The sentence 'interactions are scaled by ( n−/2)' should read 'scaled by n^{-1/2}'.
  3. [Figure captions, Figures 4.1–4.4] The parameter notation in the captions is garbled: '=e = 1' should be τ_e = 1, '<e = 1' should be σ_e = 1, and similarly for inhibitory parameters. In Figure 4.1 the caption also lists 'Gii(z) = z/2' while the text sets Gii(z)=C_ii z with C_ii=0.5; this is a typo.
  4. [Corollary 3.3, Eq. (3.9)] The constants c_{αe,ab} and c_{αi,ab} are used in (3.9) without restating their relation to the coefficients c_{αβ,ij} introduced in Hypothesis 2.2; a brief explanation of the notational correspondence would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the kinetic limits are derived from the particle SDEs, and the balance constraint is an imposed condition, not a fitted parameter.

full rationale

Walking the derivation chain from the particle system (2.5)-(2.6) through the fluctuation decomposition (2.18)-(2.21), the balance map G (2.9), and the limiting SDEs (3.2)-(3.3), I find no step in which a claimed prediction reduces by construction to an input or to a load-bearing self-citation. The limiting means v_bar are not fitted: they are defined implicitly by the balance constraint G(v_bar, mu_t) = 0 and by the differentiated dynamics in Corollary 3.3, which are derived from the particle equations via Itô's formula. The Fokker-Planck equation (3.11) is the standard generator equation for (3.2)-(3.3), not a renamed empirical quantity. The numerical tests in Section 4 set parameters a priori and compare analytic ODE solutions to simulations; no constant is calibrated to force agreement. The only self-citation, [44], is used as a literature comparison ('In an earlier paper [44] we determined the kinetic limit...'), not as the engine of existence or uniqueness; Lemma 3.1's uniqueness is proved in Section 6.1 by a Lipschitz contraction and Picard iteration. Section 4.3 explicitly reports simulations where the balance condition fails and oscillations appear, which further confirms that the theorem's conclusions are not assumed by construction. One caveat is a possible internal proof gap: Lemma 6.4 asserts that the bar y^j processes are independent and invokes the Law of Large Numbers, while (6.14)-(6.15) contain a common empirical mean-field term. That concern, if valid, undermines a proof step, but it is not circularity, because the limiting object does not presuppose the convergence it is meant to establish. Under the stated hypotheses, the theorem is a genuine n-to-infinity statement with no fitted input or definitional equivalence.

Assumptions & free parameters 0 free parameters · 7 assumptions · 0 invented entities

The analysis introduces no new physical entities or fitted constants. The central theorem is conditional on the stated hypotheses, the most consequential being the spectral gap of the balance Jacobian and the initial condition on the balanced manifold. The numerical examples use parameter values only as illustrations.

assumptions (7)
  • domain assumption Hypothesis 2.1: empirical spatial distribution converges to kappa
    Ensures the network geometry has a well-defined thermodynamic limit; used throughout Section 2.
  • domain assumption Hypothesis 2.2: connectivity kernels are finite rank with orthonormal basis functions
    Reduces the mean activity to a finite 2M-dimensional system; required for the ODE for v in Corollary 3.3.
  • domain assumption Hypothesis 2.4: f, G, sigma are C^2 with bounded derivatives up to second order; sigma is bounded
    Guarantees strong solutions, Ito formula, and the Lipschitz estimates in Lemmas 6.1-6.5.
  • domain assumption Hypothesis 2.5: initial empirical measure converges to mu0, with L2 density, finite second moments, and (v_bar(0), mu0) in B
    The theorem only starts from balance; this is a load-bearing restriction.
  • domain assumption Balanced manifold B requires every eigenvalue of J(v, mu) to have strictly negative real part
    Provides the exponential attraction that controls the n^{-1/2} interaction terms; without it the limiting result does not hold.
  • domain assumption n is large enough that det(Q) > 1/2 (equation 2.17)
    Technical assumption ensuring the projection Q is invertible for large n.
  • standard math Standard SDE existence and uniqueness theory, Ito calculus, and the Law of Large Numbers
    Used for existence and uniqueness of strong solutions (Lemma 6.3) and convergence of i.i.d. empirical measures (Lemma 6.4).

how reviews work

0 comments
Cite this review

Pith. "Pith review of The Kinetic Limit of Balanced Neural Networks." pith.science (2026). https://pith.science/paper/FNM7CRRO

@misc{pith2026250518481,
  author       = {Pith},
  title        = {Pith review of: The Kinetic Limit of Balanced Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FNM7CRRO}},
  note         = {Machine review of arXiv:2505.18481}
}
read the original abstract

The theory of Balanced Neural Networks is a very popular explanation for the high degree of variability and stochasticity in the brain's activity. Roughly speaking, it entails that typical neurons receive many excitatory and inhibitory inputs. The network-wide mean inputs cancel, and one is left with the stochastic fluctuations about the mean. In this paper we determine kinetic equations that describe the population density. The intrinsic dynamics is nonlinear, with multiplicative noise perturbing the state of each neuron. The equations have a spatial dimension, such that the strength-of-connection between neurons is a function of their spatial position. Our method of proof is to decompose the state variables into (i) the network-wide average activity, and (ii) fluctuations about this mean. In the limit, we determine two coupled limiting equations. The requirement that the system be balanced yields implicit equations for the evolution of the average activity. In the large n limit, the population density of the fluctuations evolves according to a Fokker-Planck equation. If one makes an additional assumption that the intrinsic dynamics is linear and the noise is not multiplicative, then one obtains a spatially-distributed neural field equation.

Figures

Figures reproduced from arXiv: 2505.18481 by the authors.

Figure 4
Figure 4. [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 4
Figure 4. [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 4
Figure 4. [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4 [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

67 extracted references · 66 canonical work pages

  1. [44]

    The hydrodynamic limit of balanced neural networks

    James MacLaurin and Pedro Vilanova. The hydrodynamic limit of balanced neural networks. Under Review In SIAM Applied Analysis , 2025

    work page 2025

  2. [1]

    Adams and James MacLaurin

    Zachary P. Adams and James MacLaurin. The isochronal phase of stochastic pde and integral equations: Metastability and other properties. Journal of Differential Equations , 414:773–816, 1 2025

    work page 2025

  3. [2]

    Multivariate hawkes processes on inhomogeneous random graphs

    Zo´ e Agathe-Nerine. Multivariate hawkes processes on inhomogeneous random graphs. Stochastic Processes and their Applications , 152:86–148, 10 2022

    work page 2022

  4. [3]

    Large deviations theory for markov jump models of chemical reaction networks

    Andrea Agazzi, Amir Dembo, and Jean Pierre Eckmann. Large deviations theory for markov jump models of chemical reaction networks. Annals of Applied Probability , 28:1821–1855, 2018

    work page 2018

  5. [4]

    Model of global spontaneous activity and local structured activity during delay periods in the cerebral cortex

    Daniel Amit and Nicolas Brunel. Model of global spontaneous activity and local structured activity during delay periods in the cerebral cortex. Cerebral Cortex, 7:237–252, 1997

    work page 1997

  6. [5]

    Stochastic Analysis of Biochemical Systems

    David F Anderson and Thomas G Kurtz. Stochastic Analysis of Biochemical Systems . Springer, 2015

    work page 2015

  7. [6]

    Large deviations for langevin spin glass dynamics

    G Ben Arous and A Guionnet. Large deviations for langevin spin glass dynamics. Probability Theory and Related Fields, 102, 1995

    work page 1995

  8. [7]

    Cugliandolo-kurchan equations for dynamics of spin- glasses

    Gerard Ben Arous, Amir Dembo, and Alice Guionnet. Cugliandolo-kurchan equations for dynamics of spin- glasses. Probability Theory and Related Fields , 136:619–660, 2006

    work page 2006

Show all 67 references
  1. [8]

    Neural fields and noise-induced patterns in neurons on large disordered networks

    Daniele Avitabile and James Maclaurin. Neural fields and noise-induced patterns in neurons on large disordered networks. Arxiv 2408.12540v1, 2024

    2024 arXiv

  2. [9]

    Mean-field description and propa- gation of chaos in networks of hodgkin-huxley and fitzhugh-nagumo neurons

    Javier Baladron, Diego Fasoli, Olivier Faugeras, and Jonathan Touboul. Mean-field description and propa- gation of chaos in networks of hodgkin-huxley and fitzhugh-nagumo neurons. Journal of Mathematical Neuroscience, 2:1–67, 2012

    2012

  3. [10]

    Barranca, Asha Bhuiyan, Max Sundgren, and Fangzhou Xing

    Victor J. Barranca, Asha Bhuiyan, Max Sundgren, and Fangzhou Xing. Functional implications of dale’s law in balanced neuronal network dynamics and decision making. Frontiers in Neuroscience, 16, 2 2022

    2022

  4. [11]

    Noise-Induced Phenomena in Slow-Fast Dynamical Systems

    Nils Berglund and Barbara Gentz. Noise-Induced Phenomena in Slow-Fast Dynamical Systems . Springer, 2006

    2006

  5. [12]

    Machens, and Sophie Den` eve

    Martin Boerlin, Christian K. Machens, and Sophie Den` eve. Predictive coding of dynamical variables in balanced spiking networks. PLoS Computational Biology , 9, 2013

    2013

  6. [13]

    Pattern formation in random networks using graphons

    Jason Bramburger and Matt Holzer. Pattern formation in random networks using graphons. SIAM Journal on Mathematical Analysis , 55:2150–2185, 2023

    2023

  7. [14]

    Bressloff

    Paul C. Bressloff. Spatiotemporal dynamics of continuum neural fields. Journal of Physics A: Mathematical and Theoretical, 45, 2012

    2012

  8. [15]

    Dynamics of sparsely connected networks of excitatory and inhibitory spiking neurons

    Nicolas Brunel. Dynamics of sparsely connected networks of excitatory and inhibitory spiking neurons. Journal of Computational Neuroscience , 8:183–208, 2000

    2000

  9. [16]

    Effects of neuromodulation in a cortical network model of object working memory dominated by recurrent inhibition

    Nicolas Brunel and Xiao-Jing Wang. Effects of neuromodulation in a cortical network model of object working memory dominated by recurrent inhibition. Technical report, 2001

    2001

  10. [17]

    Propagation of chaos: a review of models, methods and applications

    Louis-Pierre Chaintron and Antoine Diez. Propagation of chaos: a review of models, methods and applications. i. models and methods. Kinetic and Related Models , 15, 2 2022

    2022

  11. [18]

    Spin Glass Theory and Far Beyond

    Patrick Charbonneau, Enzo Marinari, Mark Mezard, Giorgio Parisi, Federico Ricci-Tersenghi, Gabriella Sicuro, and Francesco Zamponi, editors. Spin Glass Theory and Far Beyond . World Scientific, 2023

    2023

  12. [19]

    Mean-field limit of generalized hawkes processes

    Julien Chevallier. Mean-field limit of generalized hawkes processes. Stochastic Processes and their Applications, 127:3870–3912, 2017

    2017

  13. [20]

    Cook, Andre D

    Blake J. Cook, Andre D. H. Peterson, Wessel Woldman, and John R. Terry. Neural field models: A math- ematical overview and unifying framework. Mathematical Neuroscience and Applications , Volume 2, 3 2022. 24

    2022

  14. [21]

    Neural Fields: Theory and Applications

    Stephen Coombes, Peter beim Graben, Roland Potthast, and James Wright. Neural Fields: Theory and Applications. Springer, 2014

    2014

  15. [22]

    Crisanti and H

    A. Crisanti and H. Sompolinsky. Path integral approach to random neural networks. Physical Review E , 98:1–20, 2018

    2018

  16. [23]

    Andrea Crisanti, Heinz Horner, and H.J. Sommers. The spherical p-spin interaction spin-glass model. Zeit- schrift fur Physik B Condensed Matter , 92:257–271, 1993

    1993

  17. [24]

    Strength of correlations in strongly recurrent neuronal networks

    Ran Darshan, Carl Van Vreeswijk, and David Hansel. Strength of correlations in strongly recurrent neuronal networks. Physical Review X , 8:31072, 2018

    2018

  18. [25]

    Interacting hawkes processes with multiplicative inhibition

    C´ eline Duval, Eric Lu¸ con, and Christophe Pouzat. Interacting hawkes processes with multiplicative inhibition. Technical report, 2022

    2022

  19. [26]

    Spatially extended balanced networks without translationally invariant connectivity

    Christopher Ebsch and Robert Rosenbaum. Spatially extended balanced networks without translationally invariant connectivity. Journal of Mathematical Neuroscience , 10, 2020

    2020

  20. [27]

    Rainer Engelken, Alessandro Ingrosso, Ramin Khajeh, and Sven Goedeke L.F. Abbott. Input correlations impede suppression of chaos and learning in balanced firing-rate networks. PLoS Computational Biology, 18, 12 2022

    2022

  21. [28]

    Neural networks as spatio-temporal pattern-forming systems

    Bard Ermentrout. Neural networks as spatio-temporal pattern-forming systems. Reports on progress in physics, 61(4):353, 1998

    1998

  22. [29]

    Mean field limits for interacting hawkes processes in a diffusive regime

    Xavier Erny, Eva L¨ ocherbach, and Dasha Loukianova. Mean field limits for interacting hawkes processes in a diffusive regime. Bernoulli, 28:125–149, 2 2021

    2021

  23. [30]

    Asymptotic description of neural networks with correlated synaptic weights

    Olivier Faugeras and James MacLaurin. Asymptotic description of neural networks with correlated synaptic weights. Entropy, 17:4701–4743, 2015

    2015

  24. [31]

    Mechanisms underlying gain modulation in the cortex

    Katie A Ferguson and Jessica A Cardin. Mechanisms underlying gain modulation in the cortex. Nature Reviews Neuroscience, 21(2):80–92, 2020

    2020

  25. [32]

    Asynchronous states and the emergence of synchrony in large networks of interacting excitatory and inhibitory neurons

    D Hansel and G Mato. Asynchronous states and the emergence of synchrony in large networks of interacting excitatory and inhibitory neurons. Neural Computation, 15:1–56, 2003

    2003

  26. [33]

    Statistical field theory for neural networks

    Moritz Helias and David Dahmen. Statistical field theory for neural networks. pages 1–107, 2019

    2019

  27. [34]

    Pattern formation in intracortical neuronal fields

    Axel Hutt, Michael Bestehorn, and Thomas Wennekers. Pattern formation in intracortical neuronal fields. Network: Computation in Neural Systems , 14(2):351, 2003

    2003

  28. [35]

    Mean field limit for stochastic particle systems

    Pierre-Emmanuel Jabin and Zhenfu Wang. Mean field limit for stochastic particle systems. Active Particles, Volume 1: Advances in Theory, Models, and Applications , 2017

    2017

  29. [36]

    Predictive coding in balanced neural networks with noise, chaos and delays

    Jonathan Kadmon, Jonathan Timcheck, and Surya Ganguli. Predictive coding in balanced neural networks with noise, chaos and delays . 2020

    2020

  30. [37]

    Brownian Motion and Stochastic Calculus 2nd Edition

    Ioannis Karatzas and Steven Shreve. Brownian Motion and Stochastic Calculus 2nd Edition . Springer, 1991

    1991

  31. [38]

    Solutions of a stochastic differential equation forced onto a manifold by a large drift

    G.S Katzenberger. Solutions of a stochastic differential equation forced onto a manifold by a large drift. The Annals of Probability, 19:1587–1628, 1991

    1991

  32. [39]

    Sukbin Lim and Mark S. Goldman. Balanced cortical microcircuitry for spatial working memory based on corrective feedback control. Journal of Neuroscience , 34:6790–6806, 2014

    2014

  33. [40]

    Slow dynamics and high variability in balanced cortical networks with clustered connections

    Ashok Litwin-Kumar and Brent Doiron. Slow dynamics and high variability in balanced cortical networks with clustered connections. Nature Neuroscience, 15:1498–1505, 2012

    2012

  34. [41]

    Quenched asymptotics for interacting diffusions on inhomogeneous random graphs

    Eric Lucon. Quenched asymptotics for interacting diffusions on inhomogeneous random graphs. Stochastic Processes and their Applications , 130:6783–6842, 11 2020

    2020

  35. [42]

    Large deviations of hawkes processes on structured sparse random graphs

    J Maclaurin. Large deviations of hawkes processes on structured sparse random graphs. Under Review in Stochastic Processes and their Applications , 2024

    2024

  36. [43]

    Population-level activity in large random neural networks

    James MacLaurin, Moshe Silverstein, and Pedro Vilanova. Population-level activity in large random neural networks. Under Review in Annals of Applied Probability , 2024

    2024

  37. [45]

    De Masi, A

    A. De Masi, A. Galves, E. L¨ ocherbach, and E. Presutti. Hydrodynamic limit for interacting neurons. Journal of Statistical Physics , 158:866–902, 2014

    2014

  38. [46]

    Dynamical entropy production in spiking neuron networks in the balanced state

    Michael Monteforte and Fred Wolf. Dynamical entropy production in spiking neuron networks in the balanced state. Physical Review Letters, 105, 12 2010

    2010

  39. [47]

    Large deviations and mean-field theory for asymmetric random recurrent neural networks

    Olivier Moynot and Manuel Samuelides. Large deviations and mean-field theory for asymmetric random recurrent neural networks. Probability Theory and Related Fields , 123:41–75, 5 2002

    2002

  40. [48]

    Parsons and Tim Rogers

    Todd L. Parsons and Tim Rogers. Dimension reduction for stochastic dynamical systems forced onto a manifold by large drift: A constructive approach with examples from theoretical biology. Journal of Physics A: Mathematical and Theoretical, 50, 2017

    2017

  41. [49]

    Selectivity and sparseness in randomly connected balanced networks

    Cengiz Pehlevan and Haim Sompolinsky. Selectivity and sparseness in randomly connected balanced networks. PLoS ONE, 9, 2 2014

    2014

  42. [50]

    Pfaffelhuber, S

    P. Pfaffelhuber, S. Rotter, and J. Stiefel. Mean-field limits for non-linear hawkes processes with excitation and inhibition. Stochastic Processes and their Applications , 153:57–78, 11 2022

    2022

  43. [51]

    Balanced dynamics in strongly coupled networks

    Cristobal Quininao and Jonathan Touboul. Balanced dynamics in strongly coupled networks. arXiv preprint arXiv:2501.11769, 2025

    2025

  44. [52]

    Balanced networks of spiking neurons with spatially dependent recurrent connections

    Robert Rosenbaum and Brent Doiron. Balanced networks of spiking neurons with spatially dependent recurrent connections. Physical Review X , pages 1–9, 2014

    2014

  45. [53]

    Smith, Adam Kohn, Jonathan E

    Robert Rosenbaum, Matthew A. Smith, Adam Kohn, Jonathan E. Rubin, and Brent Doiron. The spatial 25 structure of correlated neuronal variability. Nature Neuroscience, 20:107–114, 2017

    2017

  46. [54]

    Ran Rubin, L. F. Abbott, and Haim Sompolinsky. Balanced excitation and inhibition are required for high- capacity, noise-robust neuronal selectivity. Proceedings of the National Academy of Sciences of the United States of America , 114:E9366–E9375, 2017

    2017

  47. [55]

    Noise, neural codes and cortical organization

    Michael N Shadlen and William T Newsome. Noise, neural codes and cortical organization. Current Opinion in Neurobiology, 4:569–579, 1994

    1994

  48. [56]

    The variable discharge of cortical neurons : Implications for connectivity , computation , and information coding

    Michael N Shadlen and William T Newsome. The variable discharge of cortical neurons : Implications for connectivity , computation , and information coding. The Journal of Neuroscience , 18:3870–3896, 1998

    1998

  49. [57]

    The highly irregular firing of cortical cells is inconsistent with temporal integration of random epsps

    William Softky and Christof Koch. The highly irregular firing of cortical cells is inconsistent with temporal integration of random epsps. The Journal of Neuroscience , 13:334–350, 1993

    1993

  50. [58]

    Time-dependent order parameters in spin-glasses

    H Sompolinsky. Time-dependent order parameters in spin-glasses. Physical Review Letters, 47:935–938, 1981

    1981

  51. [59]

    Topics in Propagation of Chaos

    Alain-Sol Sznitman. Topics in Propagation of Chaos . Springer-Verlag, 1989

    1989

  52. [60]

    Bistability and up / down state alternations in inhibition- dominated randomly connected networks of lif neurons

    Elisa M Tartaglia and Nicolas Brunel. Bistability and up / down state alternations in inhibition- dominated randomly connected networks of lif neurons. Scientific Reports, pages 1–14, 2017

    2017

  53. [61]

    Chaotic neural dynamics facilitate probabilistic computations through sam- pling

    Yu Terada and Taro Toyoizumi. Chaotic neural dynamics facilitate probabilistic computations through sam- pling. 2024

    2024

  54. [62]

    M. V. Tsodyks and T. Sejnowski. Rapid state switching in balanced cortical network models. Network: Computation in Neural Systems , 6:111–124, 1995

    1995

  55. [63]

    Paradoxical effects of external modulation of inhibitory interneurons

    Misha V Tsodyks, William E Skaggs, Terrence J Sejnowski, and Bruce L McNaughton. Paradoxical effects of external modulation of inhibitory interneurons. Journal of neuroscience , 17(11):4382–4388, 1997

    1997

  56. [64]

    van Vreeswijk and H

    C. van Vreeswijk and H. Sompolinsky. Chaos in neuronal networks with balanced excitatory and inhibitory. Science, 274:1724–1726, 1996

    1996

  57. [65]

    Chaos in neuronal networks with balanced excitatory and inhibitory activity

    C Van Vreeswijk and H Sompolinsky. Chaos in neuronal networks with balanced excitatory and inhibitory activity. Science, 274:1724–1726, 1996

    1996

  58. [66]

    Van Vreeswijk and H

    C. Van Vreeswijk and H. Sompolinsky. Chaotic balanced state in a model of cortical circuits. Neural Compu- tation, 10:1321–1371, 1998

    1998

  59. [67]

    Irregular Activity in Large Networks of Neurons

    Carl Van Vreeswijk and Haim Sompolinsky. Irregular Activity in Large Networks of Neurons . Elsevier, 2003. 26

    2003

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.