Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Mean-field analysis of a neural network with stochastic STDP

T0 review · 4 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read A network of binary spiking neurons with stochastic STDP has a mean-field limit described by a single typical neuron whose distribution obeys a McKean-Vlasov PDE–jump system.

desk verdict A genuinely new mean-field reduction for asymmetric stochastic STDP, but the central limit is a conjecture, not a theorem, and the abstract overclaims. read the letter →

arxiv 2510.02545 v1 pith:SNHF4LGB submitted 2025-10-02 physics.bio-ph cond-mat.dis-nnmath.PRq-bio.NC

Pascal Helson , Etienne Tanré , Romain Veltz This is my paper · ORCID
classification physics.bio-phcond-mat.dis-nnmath.PRq-bio.NC MSC 60K3592B2060J25
keywords mean-fieldlimitspike-timing-dependentplasticityWilson-CowanmodelMcKean-VlasovequationpiecewisedeterministicMarkovprocesstypicalneuronsynapticspikingneuralnetwork
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

The paper establishes (as a conjectured limit under assumptions A1–A4) that a network of N all-to-all connected binary spiking neurons with stochastic spike-timing-dependent plasticity possesses a mean-field description in terms of a single 'typical neuron': its binary activity state, the time since its last spike, and the empirical distribution of its pre-synaptic partners. The limit object is a McKean–Vlasov piecewise deterministic Markov process, in which the typical neuron's dynamics depend on its own distribution, and the distribution evolves by a transport PDE with boundary conditions and a jump-induced mass transfer. The authors argue this is the first mean-field limit for an STDP network that does not require plasticity to be slow relative to firing or the STDP curve to be symmetric, and they provide numerical evidence that the limit tracks the finite network's averages and distributions. If established rigorously, the result would cut simulation cost from order N² to order N and give a tractable object for studying how plasticity reshapes network dynamics.

What carries the argument

The central object is the 'typical neuron' X = (V, S, ξ) living on the state space {0,1} × R+ × P({0,1} × R+ × Z). The time-since-last-spike S doubles as a label: because the S values of distinct neurons are almost surely distinct (Lemma 8), each atom of a neuron's ξ can be uniquely identified with a pre-synaptic neuron, which is what makes the empirical measure dynamics Markov. The paper's calculus revolves around the Fréchet derivative of test functions with respect to ξ; jumps of ξ of magnitude 1/N are invisible at the level of the measure μ but produce drift terms in the limit PDE. The potentiation map ν+ (which replaces each weight atom w by a mixture of w and w+1 with probabilities p+(

What would settle it

Simulate the all-to-all network (parameters of Section III) with increasing N, and at a fixed time t = 500 ms compute the empirical measure of the synaptic currents I^N; if its distance to the mean-field prediction does not tend to zero as N grows, the conjecture is false. A sharper test is to compute the alleged limit functions in equations (S14)/(S23) from the N-particle data and check whether they converge; the paper explicitly leaves this continuity step unproved.

Watch

Extended reading notes

Core claim

Main Result 5 states that, under Assumptions A1–A4, the empirical distribution μ*_t of the network is the law of a typical neuron X*_t = (V*_t, S*_t, ξ*_t). The binary activity V* toggles from 0 to 1 at rate α(I(ξ*)) (with I the mean synaptic current evaluated from ξ*) and back to 0 at rate β; S* grows linearly between spikes and resets to 0 at each 0→1 jump; and ξ*, a probability distribution over the pre-synaptic triplets (V, S, W), solves a PDE: transport in S, mass exchange between V=0 and V=1 sectors at rates β and the firing-rate term, reset boundary conditions at S=0, and a potentiation operator ν+ that shifts weight mass upward at the typical neuron's own spikes. The paper derives th

Load-bearing premise

The entire derivation rests on Assumption A2: that the empirical measures of the finite system converge in distribution to a single deterministic measure as N grows; if that convergence fails, the McKean–Vlasov PDE/PDMP is just a formal guess, not the network's actual limit.

Editorial extensions

If this is right

  • If the conjecture holds, simulating the mean-field system costs O(N) instead of O(N²) for the all-to-all network, making parameter sweeps feasible.
  • The limit reproduces not just mean activity but full distributions—e.g., the wide distribution of synaptic currents that emerges from a Dirac initial condition (Figure 1d).
  • The framework applies without slow-plasticity or symmetric-STDP assumptions, which the paper cites as conflicting with experimental observations of plasticity timescales.
  • The PDE/PDMP system is a concrete object on which to study long-time behavior, weight divergence, and the effect of stimulation protocols (e.g., deep brain stimulation).
  • The construction extends in principle to other models (Dale's law, different membrane dynamics) because the assumptions on α and p± are weak.

Reading between the lines

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

  • A rigorous propagation-of-chaos proof would need to establish the continuity of the maps (S14)/(S23) that the paper leaves as an expectation; if that continuity fails, the derived system is only a formal candidate. This is the most direct path to upgrading the conjecture to a theorem.
  • The 'typical neuron' closure is essentially a mean-field ansatz for the empirical measure of a measure-valued process; it may transfer to other plastic interacting-particle systems (Ising models with adaptive couplings, epidemic networks), as the authors hint.
  • The paper's numerics discretize the S-axis and bound it (Ms = 15 ms); a testable extension is to push the bound and grid finer to see whether the small shifts seen in Figure 1b/e vanish, which would confirm those shifts are pure approximation error.
  • Because the MKV-MF tracks transients, it could serve as a fast surrogate for exploring stimulation protocols in models of plasticity-driven memory formation, though the paper does not itself perform such an exploration.
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

4 major / 5 minor

Summary. The paper studies an all-to-all network of N stochastic binary neurons with spike-timing-dependent plasticity (STDP), where synaptic weights evolve by bounded jump probabilities p± depending on the time since the last spike. The authors introduce an enriched 'typical neuron' X = (V, S, ξ), where ξ is the empirical distribution of the triplets (V, S, W) of pre-synaptic neurons. The main result (Main Result 5) claims that, as N → ∞, the empirical distribution of such typical neurons converges to a deterministic McKean–Vlasov limit: a PDMP with S increasing linearly, V toggling at rates α(I(ξ*)) and β, and ξ* solving a PDE with boundary conditions involving density ratios and the potentiation map ν+. The derivation is based on the generator of the empirical measure (µ^N_t), with formal limit passages for the drift and jump terms. Numerical simulations compare a finite network (N = 5000) with a simulation of N independent copies of the candidate limit equations, showing good qualitative and quantitative agreement for the chosen parameters.

Significance. If the identified limit is established rigorously, this would be a valuable contribution: it would provide the first mean-field description of a spiking network with STDP that does not rely on slow-fast separation, symmetric STDP curves, or Ott–Antonsen-type ansätze. The framework keeps synaptic weight heterogeneity and spike correlations encoded in the time-since-last-spike variable, and the numerical scheme is O(N) instead of O(N²) for the original network. The paper also contains a careful, self-consistent formal computation of the generator of the empirical measure, and the numerical comparison is conducted without tuning parameters to force agreement. However, the central convergence statement is not proven: Assumption A2 postulates the propagation of chaos, and the passage to the limit through non-continuous terms is explicitly left open. The paper itself labels the limit derivation a conjecture and lists tightness and uniqueness as future work. Thus the contribution is best assessed as a formal mean-field candidate with promising numerical evidence, not a proven limit theorem.

major comments (4)
  1. [Supplementary S1.B, Assumption A2; Main Result 5] Assumption A2 states that the empirical measures µ^N converge in distribution to a deterministic limit µ*. This is exactly the propagation-of-chaos statement that the paper needs to establish. The rest of the derivation, including Main Result 5, rests on this assumption. Since no proof of this convergence is provided, Main Result 5 is a conjecture rather than a theorem. The abstract's claim that the model is 'mathematically exact' is therefore unsupported. The paper should either prove a version of A2 under explicit conditions or explicitly frame Main Result 5 as a conjectured limit, with the numerical experiments as supporting evidence.
  2. [Supplementary S1.C, after Proposition 19, and S1.D, after Proposition 23] The convergence of the key terms (S14) and (S23) is not established. These terms involve the Radon–Nikodym density ratios γ_ξ = dξ(0,·,w)/dξ(0,·,Z), which are not weakly continuous functions of ξ. Consequently, weak convergence of µ^N to µ* does not automatically imply convergence of the integrals defining ν_0 and ν_−,1. The paper acknowledges that the maps are 'not a priori continuous' and that convergence is 'expected' via an unspecified approximation. This is a load-bearing gap because these terms appear in the boundary conditions and drift of the PDE for ξ* in Main Result 5. Without a concrete continuity argument or a counterexample showing the assumption is needed, the derivation stops at the level of a formal limit.
  3. [Main Result 5 and Section IV Discussion] The claimed status of Main Result 5 conflicts with the paper's own statements. Conjecture 20 is explicitly called a conjecture; Section S1.D ends with 'we should get' the limit dynamics; and the Discussion lists proving tightness and uniqueness as 'next steps.' The abstract nevertheless states the model is 'mathematically exact.' This overstates the result. The authors should either provide the missing convergence proof or rewrite the abstract, Main Result, and conclusion to clearly state that the PDE/PDMP system is a conjectured mean-field limit, valid under unproved assumptions A2 and A4.
  4. [Section III, Figure 1 and Supplementary S2] The numerical evidence would be more convincing with error bars or multiple independent realizations. The current comparison is between one finite-network simulation and one simulation of the candidate limit system. Additionally, the mean-field simulation itself uses an approximate empirical measure µ*^N rather than the actual µ* and replaces the conditional firing rate a^0_t(s) by a polynomial fit of order 5 (S2.C.2). These approximations are sensible, but they mean that any discrepancy visible in Figure 1—including the long-time shift in panel (b)—cannot be cleanly attributed to finite-N effects versus discretization error. Quantifying these two sources of error would strengthen the claim that the limit system tracks the finite network accurately.
minor comments (5)
  1. [Introduction, 'STPD' typo] In the paragraph beginning 'The proposed framework is not limited to the study of STPD', 'STPD' should be 'STDP'.
  2. [Section I.A, equation (1)] The diagonal weights W^{ii,N} are said not to affect dynamics because V^i=0 when spiking. It may help to state explicitly that W^{ii} is excluded from the sum or that the product W^{ii} V^i is zero in the relevant states.
  3. [Notation 15 and surrounding text] The symbol E with an equals sign above it (E=) is used for equality in expectation. This notation is nonstandard and is not defined at first use in the main text; consider using a clearer notation such as 'E[X] = E[Y]'.
  4. [Section III, Figure 1 caption] The caption refers to '1a' through '1f' but in the text these are sometimes called 'Figure 1a' and sometimes 'Figure 1(a)'. Please standardize.
  5. [Supplementary S1.B, after Assumption A4] Assumption A4 imposes a C^1 density and boundary conditions at t=0, but no existence or uniqueness of the resulting PDE system is discussed. A short remark on the well-posedness of the candidate limit equations would help the reader know what is being assumed.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the mean-field limit is derived from the finite-N generator and checked by independent simulation; the main gap is an unproved convergence assumption, not a self-referential reduction.

full rationale

The derivation is not circular. The finite-N dynamics are fully specified in Section I A, and the limit system is obtained by taking N to infinity in the generator/expectation identities (S3), (S15), (S20), with the rate terms nu_0, nu_{-1,1}, and nu_+ defined from the particle dynamics in (S11), (S22), and (S24). These terms are not fitted to the limit solution, and the numerical comparison in Section III uses the same parameters and initial conditions for both the finite network and the mean-field particle simulation, with no parameter tuned to align the curves. The load-bearing caveat is Assumption A2, which postulates convergence of the empirical measures to a deterministic mu*; together with the explicitly admitted lack of continuity of the maps in (S14)/(S23) and the authors' own wording that the limit is 'expected' and 'we should get' it, this means Main Result 5 is a conjectured limit rather than a proved theorem. But that is an unproved propagation-of-chaos statement, not a reduction of the limit equations to the assumption by construction: A2 does not specify the form of the limit, and the PDE follows from the generator rather than being inserted as an ansatz. The self-citations [12,13] merely identify the microscopic model and are not used to justify the limit passage. The abstract's phrase 'mathematically exact' overstates the conditional/conjectural status of the result, but that is a correctness/overclaim concern, not circularity.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

No parameters are fitted to the target results; the simulation constants are model inputs. The 'typical neuron' is a mathematical state variable, not a new physical entity. The heavy assumptions are A2 (convergence) and A4 (regularity), both stated in the supplementary.

assumptions (5)
  • domain assumption Assumption A1: initial neuron states (V,S) and weights are i.i.d.; S has a density with bounded density independent of N
    Required for exchangeability and for uniqueness of time-since-last-spike labels; stated in 'Assumption A1'.
  • ad hoc to paper Assumption A2: empirical distributions µ^N converge in distribution to a deterministic limit µ*
    This is the propagation-of-chaos statement the paper aims to establish; it is assumed, not proved. Without it, Main Result 5 is only a candidate limit.
  • standard math Assumption A3: ξ↦α(I(ξ)) and s↦p±(s,w) are bounded continuous
    Technical regularity needed for the generator computations; stated as Assumption A3.
  • domain assumption Assumption A4: ξ*_t admits a C^1 density in s and the boundary conditions hold at t=0
    Needed to write the PDE and integrate by parts; the boundary value ρ1(0) is imposed via equation (S27).
  • standard math Test functions Ψ are C^{1,1}_b (differentiable in s, Fréchet differentiable in ξ)
    Functional setting for deriving the generator; Notation 13.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mean-field analysis of a neural network with stochastic STDP." pith.science (2026). https://pith.science/paper/SNHF4LGB

@misc{pith2026251002545,
  author       = {Pith},
  title        = {Pith review of: Mean-field analysis of a neural network with stochastic STDP},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SNHF4LGB}},
  note         = {Machine review of arXiv:2510.02545}
}
abstract

Analysing biological spiking neural network models with synaptic plasticity has proven to be challenging both theoretically and numerically. In a network with N all-to-all connected neurons, the number of synaptic connections is on the order of $N^2$, making these models computationally demanding. Furthermore, the intricate coupling between neuron and synapse dynamics, along with the heterogeneity generated by plasticity, hinder the use of classic theoretical tools such as mean-field or slow-fast analyses. To address these challenges, we introduce a new variable which we term a typical neuron X. Viewed as a post-synaptic neuron, X is composed of the activity state V , the time since its last spike S, and the empirical distribution $\xi$ of the triplet V , S and W (incoming weight) associated to the pre-synaptic neurons. In particular, we study a stochastic spike-timing-dependent plasticity (STDP) model of connection in a probabilistic Wilson-Cowan spiking neural network model, which features binary neural activity. Taking the large N limit, we obtain from the empirical distribution of the typical neuron a simplified yet accurate representation of the original spiking network. This mean-field limit is a piecewise deterministic Markov process (PDMP) of McKean-Vlasov type, where the typical neuron dynamics depends on its own distribution. We term this analysis McKean-Vlasov mean-field (MKV-MF). Our approach not only reduces computational complexity but also provides insights into the dynamics of this spiking neural network with plasticity. The model obtained is mathematically exact and capable of tracking transient changes. This analysis marks the first exploration of MKV-MF dynamics in a network of spiking neurons interacting with STDP.

Figures

Figures reproduced from arXiv: 2510.02545 by the authors.

Figure 1
Figure 1. FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Mechanistic bridges from receptors to whole-brain dynamics: mean-field reductions, validity domains, and computational trade-offs

    q-bio.NC 2026-07 conditional novelty 5.0 of 10

    This review reconstructs the mathematical lineage from spiking circuits to receptor-aware whole-brain mean-field models and introduces operation- and memory-cost benchmarks showing global covariance propagation is imp...

Reference graph

Works this paper leans on

80 extracted references · 1 canonical work pages · cited by 1 Pith paper

  1. [1]

    We denote byT ∗ t the spikes times, that is the times on[0, t]whenVjumps from0to1, 3.S ∗ t =S ∗ 0 +t− P τ∈T ∗ t S∗ τ − , 4.ξ ∗ t admits a density inssuch thatξ ∗ t (v, ds, w) =ξ∗ t (v, s, w)ds(abuse of notation) and between the spikes    ∂tξ∗ t (0, s, w) =−∂sξ∗ t (0, s, w) +βξ∗ t (1, s, w)− R P(E m) α I(ξ ′) ξ∗ t (0,s,w) ξ∗ t (0,s,Z) µ∗ t (0, s, dξ′) ξ...

  2. [2]

    Hence, asNtends to infinity, the limit of the term 1 depends on the Fr´ echet derivative of Ψ

    Drift term due to the returns to resting potential When a neuron returns to its resting potential state 0, the total variation of the jumps of the empirical distributions (ξ1,N t ),· · ·,(ξN,N t ) are of order 1 N . Hence, asNtends to infinity, the limit of the term 1 depends on the Fr´ echet derivative of Ψ. The direction of this derivative describes the...

  3. [3]

    Z T 0 ⟨µN u ,Ψ⟩du # →N∞ E

    Properties ofS Under Assumption A1, the neural network possesses two important properties that are extensively used throughout this paper. First, the times from the last spike are almost surely distinct. Lemma 8.Grant Assumption A1. Consider the process X N t solution of the microscopic model described in Sec- tion I A. Then, for anyt≥0and for alli̸=j∈J1,...

  4. [4]

    We describe the generator only on the set of cylindrical functions Φ∈C b(P(E)) that is there exists Ψ∈C 1,1 b (E) such that ∀µ∈ P(E),Φ(µ) =⟨µ,Ψ⟩

    Generator of(µ N t )t≥0 We now look for the generator of the Markov process (µ N t )t≥0. We describe the generator only on the set of cylindrical functions Φ∈C b(P(E)) that is there exists Ψ∈C 1,1 b (E) such that ∀µ∈ P(E),Φ(µ) =⟨µ,Ψ⟩. We consider the first jump of (µN t )t≥0 that we denote byτ >0. We split the generator into the drift term and the jump te...

  5. [5]

    We now give some details on this limit system

    At a spike timeτ∈ T ∗ ∞,ξ ∗ τ − jumps toν +{ξ∗ τ − }where for all PDFξonE m, ν+{ξ}(v, s, w):=p +(s, w−1)ξ(v, s, w−1) + (1−p+(s, w))ξ(v, s, w). We now give some details on this limit system. The second point details the membrane potential (spiking) ac- tivity of the neuron. Note that the spiking jump depends on the distributionξ ∗ t through the functionIde...

  6. [6]

    This term is more difficult to deal with because the spiking rates (1 {V i,N t =0}α(I i,N t ))1≤i≤N depend on the actual state of the neural networkX N t

    Drift term due to action potentials AsNtends to infinity, the limit of the term 2 also gives a drift term forξ ∗. This term is more difficult to deal with because the spiking rates (1 {V i,N t =0}α(I i,N t ))1≤i≤N depend on the actual state of the neural networkX N t . Informally, the limit of the term 2 represents the following mass transport on the prob...

  7. [7]

    Justifying Main Result 5 We are now in position to detail why Main Result 5 can be conjectured. Under Assumptions A1, A2 and A3, by makingNtends to infinity in equation (S20) and using the results of Propositions 23 and 24, we should get the following dynamics ofµ ∗ t : •L(X ∗ 0 ) = lim N→∞ L(X 1,N 0 ) =µ ∗ 0 and in particularS ∗ 0 is distributed by theρ ...

  8. [8]

    Parameters First, they both have the same parameters that we list here: 25 •N: number of neurons, •α: rate function with the total synaptic current as input (jump ofVfrom 0 to 1), •β: rate of the return to the resting potential (jump ofVfrom 1 to 0), •p ±: STDP functions with time delay between spikes as input and probability of synaptic weight jump as ou...

Show all 80 references
  1. [9]

    •dS i,∗ t =dt

    Approximatingξ The neuronisatisfies the following equations •X i,∗ 0 is determined from the initial state of the microscopic model. •dS i,∗ t =dt. •ξ i,∗ t admits a density insand satisfies the following equation: ∂tξi,∗ t (0, s, w) =−∂sξi,∗ t (0, s, w) +βξi,∗ t (1, s, w)−ξi,∗...

  2. [10]

    We draw (V i,N 0 )i = (V i,∗ 0 )i asNiid random variables with binomial distribution of parameterp V

  3. [11]

    Forisuch thatV i,∗ 0 = 0, we draw (S i,N 0 )i = (Si,∗ 0 )i iid with distributionsρ 0

  4. [12]

    We draw the synaptic weights (W ij,N 0 )i,j iid with distributionp W

  5. [13]

    We define for all (m, w)∈J0, MsK×Jw min, wmaxK, ξi,∗ 0 (0, mdt, w) =(1−p V )ρ0(mdt)pW (w) dt MsP ˜m=0 ρ0( ˜mdt) , whereρ 0 is distributed as a LogNormal distribution with parametersµ= 0.8 andσ= 1

  6. [14]

    We illustrate in Fig

    We first compute from (S27) and then we define for all (m, w)∈J0, M sK×Jw min, wmaxK, ξi,∗ 0 (1, mdt, w) =pV ρ1(mdt)pW (w) dt MsP ˜m=0 ρ1( ˜mdt) , whereρ 1 is an exponential distribution with parameterρ 1(0). We illustrate in Fig. S1 these initial densities that we used in our...

  7. [15]

    We draw the jumping timesτ i,N t of the neurons from exponential distributions with parameter α(I i,N t )δ{V i,N t =0} +βδ {V i,N t =1}

    The microscopic system We compute the synaptic currentsI i,N t = 1 N P j W ij,N t V j,N t . We draw the jumping timesτ i,N t of the neurons from exponential distributions with parameter α(I i,N t )δ{V i,N t =0} +βδ {V i,N t =1}. Ifτ i,N t < dtand 27 (a) (b) FIG. S1: Distributi...

  8. [16]

    The mean field system We describe how to pass fromttot+dtin the mean field system, in particular we have to describe how to discretise equation (S26)

  9. [17]

    We compute theI i,∗ t (initially all the same and deterministic)

  10. [18]

    We draw the jumping timesτ i,∗ t from the exponential distributions of parameters α(I i,∗ t )1 {V i,∗ t =0} +β1 {V i,∗ t =1}

  11. [19]

    We estimate the functiona 0 t = (a0 t (0),· · ·, a0 t (Msdt)). To do so, we can approximate this term looking back at the original term which is a0 t (mdt) = µ∗ t (0, mdt,P(Em)) ξ∗ t (0, mdt,Z) Z P(E m) α I(ξ ′) µ∗ t (0, mdt, dξ′) | {z } c .(S28) Regarding the quotient term in...

  12. [20]

    We compute the terms due to the drift of theξ. For allm∈J0, M sK: ξi,∗ t+dt(0, mdt, w) =ξi,∗ t (0,(m−1)dt, w) 1−a 0 t (m−1)dt dt +ξ i,∗ t (1,(m−1)dt, w)βdt ξi,∗ t+dt(0,0, w) = 0, ξi,∗ t+dt(1, mdt, w) =ξi,∗ t (1,(m−1)dt, w) (1−βdt) ξi,∗ t+dt(1,0, w) =dt X m a0 t (mdt) p−(Si,∗ t...

  13. [21]

    Ifτ i,∗ t < dtandV i,∗ t = 0, ξi,∗ t+dt(v, mdt, w) =p+(mdt)ξi,∗ t (v, mdt, w−1) + (1−p+(mdt))ξi,∗ t (v, mdt, w)

  14. [22]

    We update Si,∗ t+dt = (Si,∗ t +dt)1 {τ i,∗ t ≥dt} + (dt−τ i,∗ t )1 {τ i,∗ t <dt},

  15. [23]

    Forisuch thatτ i,∗ t < dt, we make the membrane potentialsV i,∗ jump accordingly

  16. [24]

    We compute the elements that we keep in memory at each time step : expectations of the potentials, of the time from last the spike, of the pre-synaptic weights, distribution of the synaptic current input. 29

  17. [25]

    G. K. Ocker, A. Litwin-Kumar, and B. Doiron, Self- Organization of Microcircuits in Networks of Spiking Neurons with Plastic Synapses, PLOS Computational Bi- ology11, 1 (2015), publisher: Public Library of Science

    2015

  18. [26]

    Mongillo, S

    G. Mongillo, S. Rumpel, and Y. Loewenstein, Intrinsic volatility of synaptic connections — a challenge to the synaptic trace theory of memory, Current Opinion in Neurobiology46, 7 (2017)

    2017

  19. [27]

    G. K. Ocker and B. Doiron, Training and spontaneous re- inforcement of neuronal assemblies by spike timing plas- ticity, Cerebral Cortex29, 937 (2019)

    2019

  20. [28]

    Brzosko, S

    Z. Brzosko, S. B. Mierau, and O. Paulsen, Neuro- modulation of Spike-Timing-Dependent Plasticity: Past, Present, and Future, Neuron103, 563 (2019)

    2019

  21. [29]

    Kempter, W

    R. Kempter, W. Gerstner, and J. L. van Hemmen, Heb- bian learning and spiking neurons, Phys. Rev. E59, 4498 (1999)

    1999

  22. [30]

    Bi and M.-m

    G.-q. Bi and M.-m. Poo, Synaptic modifications in cul- tured hippocampal neurons: dependence on spike timing, synaptic strength, and postsynaptic cell type, Journal of neuroscience18, 10464 (1998)

    1998

  23. [31]

    Robert and G

    P. Robert and G. Vignoud, Stochastic models of neural synaptic plasticity, SIAM Journal on Applied Mathemat- ics81, 1821 (2021)

    2021

  24. [32]

    Robert and G

    P. Robert and G. Vignoud, Stochastic models of neural synaptic plasticity: A scaling approach, SIAM Journal on Applied Mathematics81, 2362 (2021)

    2021

  25. [33]

    D. H. O’Connor, G. M. Wittenberg, and S. S.-H. Wang, Graded bidirectional synaptic plasticity is composed of switch-like unitary events, Proceedings of the National Academy of Sciences of the United States of America 102, 9679 (2005)

    2005

  26. [34]

    P. A. Appleby and T. Elliott, Synaptic and temporal en- semble interpretation of Spike-Timing-Dependent Plas- ticity, Neural Computation17, 2316 (2005)

    2005

  27. [35]

    P. A. Appleby and T. Elliott, Stable competitive dy- namics emerge from multispike interactions in a stochas- tic model of Spike-Timing-Dependent Plasticity, Neural Computation18, 2414 (2006)

    2006

  28. [36]

    Helson, A new stochastic STDP Rule in a neural Net- work Model, arXiv preprint arXiv:1706.00364 (2017)

    P. Helson, A new stochastic STDP Rule in a neural Net- work Model, arXiv preprint arXiv:1706.00364 (2017)

    2017 arXiv

  29. [37]

    Helson,Plasticity in networks of spiking neurons in interaction, Ph.D

    P. Helson,Plasticity in networks of spiking neurons in interaction, Ph.D. thesis, Universit´ e Cˆ ote d’Azur (2021)

    2021

  30. [38]

    Galtier and G

    M. Galtier and G. Wainrib, Multiscale analysis of slow- fast neuronal learning models with noise, J. Math. Neu- rosci.2, Art. 13 64 (2012)

    2012

  31. [39]

    Perthame, D

    B. Perthame, D. Salort, and G. Wainrib, Distributed synaptic weights in a LIF neural network and learning rules, Phys. D353/354, 20 (2017)

    2017

  32. [40]

    Kasai, N

    H. Kasai, N. E. Ziv, H. Okazaki, S. Yagishita, and T. Toy- oizumi, Spine dynamics in the brain, mental disorders and artificial neural networks, Nature Reviews Neuro- science22, 407 (2021)

    2021

  33. [41]

    Choquet and A

    D. Choquet and A. Triller, The dynamic synapse, Neuron 80, 691 (2013)

    2013

  34. [42]

    Vasin, N

    A. Vasin, N. Sabeva, C. Torres, S. Phan, E. A. Bushong, M. H. Ellisman, and M. Bykhovskaia, Two pathways for the activity-dependent growth and differentiation of synaptic boutons in drosophila, ENeuro6(2019)

    2019

  35. [43]

    Aso and G

    Y. Aso and G. M. Rubin, Dopaminergic neurons write and update memories with cell-type-specific rules, Elife 5, e16135 (2016)

    2016

  36. [44]

    K. A. Fulton, D. Zimmerman, A. Samuel, K. Vogt, and S. R. Datta, Common principles for odour coding across vertebrates and invertebrates, Nature Reviews Neuro- science , 1 (2024)

    2024

  37. [45]

    Y. E. Rodrigues, C. M. Tigaret, H. Marie, C. O’Donnell, and R. Veltz, A stochastic model of hippocampal synap- tic plasticity with geometrical readout of enzyme dynam- ics,eLife 12, 10.7554/eLife.80152 (2023)

    2023 doi

  38. [46]

    Zenke, W

    F. Zenke, W. Gerstner, and S. Ganguli, The temporal paradox of Hebbian learning and homeostatic plasticity, Current Opinion in Neurobiology43, 166 (2017)

    2017

  39. [47]

    Lansner, F

    A. Lansner, F. Fiebig, and P. Herman, Fast hebbian plas- ticity and working memory, Current opinion in neurobi- ology83, 102809 (2023)

    2023

  40. [48]

    Masquelier, R

    T. Masquelier, R. Guyonneau, and S. J. Thorpe, Spike Timing Dependent Plasticity Finds the Start of Repeat- ing Patterns in Continuous Spike Trains, PLOS ONE3, 1 (2008), publisher: Public Library of Science

    2008

  41. [49]

    Masquelier, R

    T. Masquelier, R. Guyonneau, and S. J. Thorpe, Compet- itive STDP-based spike pattern learning, Neural Compu- tation21, 1259 (2009)

    2009

  42. [50]

    Gilson, A

    M. Gilson, A. N. Burkitt, D. B. Grayden, D. A. Thomas, and J. L. van Hemmen, Emergence of network struc- ture due to Spike-Timing-Dependent Plasticity in re- current neuronal networks v: self-organization schemes and weight dependence, Biological Cybernetics103, 365 (2010)

    2010

  43. [51]

    Gilson and T

    M. Gilson and T. Fukai, Stability versus Neuronal Spe- cialization for STDP: Long-Tail Weight Distributions Solve the Dilemma, PLOS ONE6, 1 (2011), publisher: Public Library of Science

    2011

  44. [52]

    Pechersky, G

    E. Pechersky, G. Via, and A. Yambartsev, Stochastic ising model with plastic interactions, Statistics & Prob- ability Letters123, 100 (2017)

    2017

  45. [53]

    Montbri´ o, D

    E. Montbri´ o, D. Paz´ o, and A. Roxin, Macroscopic de- scription for networks of spiking neurons, Physical Re- view X5, 021028 (2015)

    2015

  46. [54]

    De Masi, A

    A. De Masi, A. Galves, E. L¨ ocherbach, and E. Pre- sutti, Hydrodynamic limit for interacting neurons, J. Stat. Phys.158, 866 (2015)

    2015

  47. [55]

    Delarue, J

    F. Delarue, J. Inglis, S. Rubenthaler, and E. Tanr´ e, Parti- cle systems with a singular mean-field self-excitation. Ap- plication to neuronal networks, Stochastic Process. Appl. 125, 2451 (2015)

    2015

  48. [56]

    Fournier and E

    N. Fournier and E. L¨ ocherbach, On a toy model of inter- acting neurons, Ann. Inst. Henri Poincar´ e Probab. Stat. 52, 1844 (2016)

    2016

  49. [57]

    Delattre and N

    S. Delattre and N. Fournier, Statistical inference versus mean field limit for Hawkes processes, Electron. J. Stat. 10, 1223 (2016)

    2016

  50. [58]

    Chevallier, Mean-field limit of generalized Hawkes pro- cesses, Stochastic Process

    J. Chevallier, Mean-field limit of generalized Hawkes pro- cesses, Stochastic Process. Appl.127, 3870 (2017)

    2017

  51. [59]

    Farkhooi and W

    F. Farkhooi and W. Stannat, Complete mean-field the- ory for dynamics of binary recurrent networks, Physical review letters119, 208301 (2017)

    2017

  52. [60]

    Huang and C.-C

    C.-H. Huang and C.-C. K. Lin, New biophysical rate- based modeling of long-term plasticity in mean-field neuronal population models, Computers in Biology and Medicine163, 107213 (2023)

    2023

  53. [61]

    R. M. Lorenzi, A. Geminiani, Y. Zerlaut, M. De Grazia, 30 A. Destexhe, C. A. Gandini Wheeler-Kingshott, F. Palesi, C. Casellato, and E. D’Angelo, A multi- layer mean-field model of the cerebellum embedding microstructure and population-specific dynamics, PLOS Computational Bio...

    2023

  54. [62]

    Duchet, C

    B. Duchet, C. Bick, and ´A. Byrne, Mean-field ap- proximations with adaptive coupling for networks with spike-timing-dependent plasticity, Neural computation 35, 1481 (2023)

    2023

  55. [63]

    Robert and G

    P. Robert and G. Vignoud, On the spontaneous dynamics of synaptic weights in stochastic models with pair-based stdp, arXiv preprint arXiv:2111.07919 (2021)

    2021 arXiv

  56. [64]

    Sznitman, Topics in propagation of chaos, in ´Ecole d’ ´Et´ e de Probabilit´ es de Saint-Flour XIX—1989, Lecture Notes in Math., Vol

    A.-S. Sznitman, Topics in propagation of chaos, in ´Ecole d’ ´Et´ e de Probabilit´ es de Saint-Flour XIX—1989, Lecture Notes in Math., Vol. 1464 (Springer, Berlin, 1991) pp. 165–251

    1989

  57. [65]

    Dolcemascolo, A

    A. Dolcemascolo, A. Miazek, R. Veltz, F. Marino, and S. Barland, Effective low-dimensional dynamics of a mean-field coupled network of slow-fast spiking lasers, Physical Review E101, 052208 (2020)

    2020

  58. [66]

    Pastor-Satorras, C

    R. Pastor-Satorras, C. Castellano, P. Van Mieghem, and A. Vespignani, Epidemic processes in complex networks, Reviews of modern physics87, 925 (2015)

    2015

  59. [67]

    A. M. Lozano, N. Lipsman, H. Bergman, P. Brown, S. Chabardes, J. W. Chang, K. Matthews, C. C. McIn- tyre, T. E. Schlaepfer, M. Schulder,et al., Deep brain stimulation: current challenges and future directions, Na- ture Reviews Neurology15, 148 (2019)

    2019

  60. [68]

    H. P. McKean, A class of markov processes associated with nonlinear parabolic equations, Proceedings of the National Academy of Sciences56, 1907 (1966)

    1907

  61. [69]

    P. C. Bressloff, Metastable states and quasicycles in a stochastic wilson-cowan model of neuronal population dynamics, Physical Review E82, 051903 (2010)

    2010

  62. [70]

    Benayoun, J

    M. Benayoun, J. D. Cowan, W. van Drongelen, and E. Wallace, Avalanches in a Stochastic Model of Spik- ing Neurons, PLOS Computational Biology6, 1 (2010), publisher: Public Library of Science

    2010

  63. [71]

    Litwin-Kumar and B

    A. Litwin-Kumar and B. Doiron, Formation and mainte- nance of neuronal assemblies through synaptic plasticity, Nature Communications5, 1 (2014)

    2014

  64. [72]

    G. K. Ocker, Y. Hu, M. A. Buice, B. Doiron, K. Josi´ c, R. Rosenbaum, and E. Shea-Brown, From the statistics of connectivity to the statistics of spike times in neu- ronal networks, Current opinion in neurobiology46, 109 (2017)

    2017

  65. [73]

    Jabin, V

    P.-E. Jabin, V. Schmutz, and D. Zhou, Non-exchangeable networks of integrate-and-fire neurons: spatially- extended mean-field limit of the empirical measure, arXiv preprint arXiv:2409.06325 (2024)

    2024 arXiv

  66. [74]

    Cormier, E

    Q. Cormier, E. Tanr´ e, and R. Veltz, Long time behavior of a mean-field model of interacting neurons, Stochastic Process. Appl.130, 2553 (2020)

    2020

  67. [75]

    D. G. Clark and L. Abbott, Theory of coupled neuronal- synaptic dynamics, Physical Review X14, 021001 (2024)

    2024

  68. [76]

    Billingsley,Convergence of probability measures, 2nd ed., Wiley Series in Probability and Statistics: Probabil- ity and Statistics (John Wiley & Sons, Inc., New York,

    P. Billingsley,Convergence of probability measures, 2nd ed., Wiley Series in Probability and Statistics: Probabil- ity and Statistics (John Wiley & Sons, Inc., New York,

  69. [78]

    A. S. Kechris,Classical descriptive set theory, Graduate Texts in Mathematics, Vol. 156 (Springer-Verlag, New York, 1995) pp. xviii+402

    1995

  70. [79]

    J. K. Hale,Ordinary differential equations, 2nd ed. (Robert E. Krieger Publishing Co., Inc., Huntington, N.Y., 1980) pp. xvi+361

    1980

  71. [80]

    Ambrosio, N

    L. Ambrosio, N. Fusco, and D. Pallara,Functions of bounded variation and free discontinuity problems, Ox- ford Mathematical Monographs (The Clarendon Press, Oxford University Press, New York, 2000) pp. xviii+434

    2000

  72. [1999]

    x+277, a Wiley-Interscience Publication

    pp. x+277, a Wiley-Interscience Publication

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.