betaFeedback

WorksMathematics

First and second order firing statistics of large neuronal networks with community structure

Abstract

We study a discrete-time stochastic model for the spiking activity of a network of nn interacting neurons partitioned into kk communities. The interaction network is a random directed weighted graph generated by a Stochastic Block Model with k+1k+1 edge types (kk intra-community, 11 inter-community). Each edge is present independently with a probability that depends only on its type. When present, an edge carries a type-dependent deterministic weight scaled by n−1n^{-1}, which is excitatory or inhibitory depending on an independent random sign. Given the interaction network, the spiking process is a stationary Markov chain. Our results describe how the network's community structure manifests itself in the first- and second-order statistics of this chain, as n→∞n \to \infty with kk fixed. Representing these statistics as explicit functions of the random interaction network, we prove that, with high probability: (i) the vectors of mean firing activity and firing variances are approximated in supremum norm by deterministic vectors that are constant within each community, with an error of order log⁡(n)/n\sqrt{\log(n)/n}; (ii) the off-diagonal part of the 00-lagged firing covariance matrix is approximated in max norm, with an error of order log⁡(n)/n3\sqrt{\log(n)/n^3}, by a deterministic community-block matrix (constant on each block defined by a pair of communities), with block values determined by the unique solution to a k×kk \times k Stein matrix equation; (iii) in contrast, the tt-lagged covariance matrix, for every lag t≥1t \ge 1, is approximated in max norm, again with an error of order log⁡(n)/n3\sqrt{\log(n)/n^3}, by a random matrix built from two components: a random matrix capturing the interaction network and a deterministic community-block matrix. Finally, we discuss conditions on the model parameters under which these limiting statistics enable the identifiability of the community structure.

Introduction

Model and problem statement

In this paper, we consider a stochastic, discrete-time model describing the spiking activity of a network of nn interacting neurons partitioned into kk communities. The model has two components, which can be informally described as follows.

The first is a directed weighted graph with nn vertices, whose edges can be of k+1k+1 types: kk intra-community types, one for each community, and a single inter-community type. An edge is of intra-community type if it connects two neurons in the same community, and of inter-community type otherwise. Each edge in the graph carries a weight whose absolute value is the same for all edges of a given type; we refer to this common value as the intensity of the edge type. The sign of the weight, however, is independent of the edge type: edges with positive weights are called excitatory, while those with negative weights are called inhibitory. For most of our results, the graph structure and the edge signs are random. Specifically, edges are present independently of one another, with probabilities depending only on their types, as in a Stochastic Block Model [1]. Independently of the edge configuration, the edge signs are independent and identically distributed, with each sign being positive with a fixed probability. We write Θ\Theta for the matrix indicating which edges are present in the graph and HH for the matrix of signs. Hereafter, we call the pair (Θ,H)(\Theta,H) the environment and denote by A=(Aij)i,j∈[n]A=(A_{ij})_{i,j\in[n]} its associated interaction matrix. The entry AijA_{ij} represents the weight of the edge from neuron ii to neuron jj, scaled by n−1n^{-1}, if the edge is present, and is 00 otherwise.

The second component is a sequence X=(Xt)t∈NX=(X_t)_{t\in\mathbb{N}} of random vectors taking values in {0,1}n\{0,1\}^n, representing the time evolution of the spikes of the nn neurons. Writing Xi,tX_{i,t} to denote the ii-th coordinate of the random vector XtX_t, we have Xi,t=1X_{i,t}=1 if neuron ii spikes at time tt, and 00 otherwise. Conditionally on the realization of the environment (Θ,H)(\Theta,H), the sequence XX forms a Markov chain. Specifically, given the environment and the entire configuration of spikes at time t−1t-1, each neuron ii spikes independently of the others at the next time step, with probability

P(Θ,H)(Xi,t=1∣Xt−1)=λ+(ATXt−1)i,(1)\mathbb{P}_{(\Theta,H)}(X_{i,t}=1\mid X_{t-1})=\lambda+(A^{\mathsf{T}}X_{t-1})_i, \tag*{(1)}

where ATA^{\mathsf{T}} denotes the transpose of the matrix AA and λ∈[0,1]\lambda\in[0,1] represents the spontaneous firing rate, which is the same for all neurons. The second term on the right-hand side of (1) is a weighted sum of the spikes emitted at time t−1t-1 by the neurons that point to neuron ii in the graph, where the weights are given by the corresponding entries of the interaction matrix AA. Since the entries of AA are of order n−1n^{-1}, the interactions are individually weak.

Throughout our analysis, we work under conditions on the model parameters (see (14) and (15)) ensuring that the right-hand side of (1) is always between 00 and 11 for a set of realizations of the environment (Θ,H)(\Theta,H) whose probability approaches 11 as n→∞n\to\infty (this is formalized in Lemma 3.1). We refer to such realizations as admissible realizations. Conditionally on admissible realizations of the environment, the Markov chain XX admits a unique invariant measure, and we always work with its stationary version. In the sequel, we denote by P(Θ,H)\mathbb{P}_{(\Theta,H)} the law of the stationary chain XX given the environment (Θ,H)(\Theta,H), and by E(Θ,H)\mathbb{E}_{(\Theta,H)} the corresponding expectation.

In what follows, let E(Θ,H)[Xt]\mathbb{E}_{(\Theta,H)}[X_t] be the column vector of the mean firing activity at time tt and let Σ(Θ,H)(t)\Sigma^{(t)}_{(\Theta,H)} denote the tt-lagged firing covariance matrix that is defined as

Σ(Θ,H)(t):=E(Θ,H)[Xt(X0)T]−E(Θ,H)[Xt](E(Θ,H)[X0])T,t∈N.\Sigma^{(t)}_{(\Theta,H)}:=\mathbb{E}_{(\Theta,H)}[X_t(X_0)^{\mathsf{T}}]-\mathbb{E}_{(\Theta,H)}[X_t](\mathbb{E}_{(\Theta,H)}[X_0])^{\mathsf{T}},\quad t\in\mathbb{N}.

By stationarity, the mean vectors E(Θ,H)[Xt]\mathbb{E}_{(\Theta,H)}[X_t] do not depend on tt, and in the rest of the paper we refer to E(Θ,H)[X0]\mathbb{E}_{(\Theta,H)}[X_0] as the vector of mean firing rates. Note that both this vector and the lagged covariance matrices are random, depending on the realization of the environment. Also, recall that the law of this environment is determined by the partition of the population into the kk communities, the edge probabilities, the intensities of the different edge types, and the probability that an edge is excitatory.

This leads to the following related problems, which we address in this paper.

Problem statement. As the size nn of the neuronal network grows while the number kk of communities is kept fixed, how does the community structure of the environment manifest itself in the mean firing rates and in the lagged firing covariance matrices? Conversely, to what extent can the partition of the population into communities be recovered from these quantities alone?

Overview of the main results

Throughout the paper, we denote by C={C1,…,Ck}\mathcal{C} = \{C_1,\ldots,C_k\} the partition of the set [n]:={1,…,n}[n] := \{1,\ldots,n\} representing the kk communities of the entire network of neurons, whose size is nn. In our analysis, we assume that all communities have sizes proportional to the total number of neurons in the network, i.e., we suppose that ∣Cb∣=αbn≥1\lvert C_b\rvert= \alpha_b n \ge1 for every b∈[k]b \in[k], for some positive values α1,…,αk\alpha_1,\ldots,\alpha_k satisfying α1+…+αk=1\alpha_1+\ldots+\alpha_k=1. In what follows, we write ∥⋅∥∞\lVert\cdot\rVert_\infty to denote the usual supremum norm in Rn\mathbb{R}^n and KK to indicate a positive constant independent of nn.

Recall that E(Θ,H)[X0]\mathbb{E}_{(\Theta,H)}[X_0] denotes the random vector of mean firing rates, where the randomness comes from the random environment (Θ,H)(\Theta,H). In our first main result, Theorem 3.1, we show how the community structure of the environment is encoded in this random vector when nn is large. Specifically, we show that with high probability

∥E(Θ,H)[X0]−m‾∥∞≤Klog⁡(n)n,\left\lVert\mathbb{E}_{(\Theta,H)}[X_0]-\overline{m}\right\rVert_\infty\le K\sqrt{\frac{\log(n)}{n}},

where m‾\overline{m} is the nn-dimensional vector defined as

m‾:=λ∑b∈[k]ζb1Cb,(2)\overline{m}:=\lambda\sum_{b\in[k]}\zeta_b\mathbf{1}_{C_b}, \tag*{(2)}

where ζ=(ζ1,…,ζk)\zeta=(\zeta_1,\ldots,\zeta_k) is an explicitly known kk-dimensional vector (see, (17)) and 1Cb\mathbf{1}_{C_b} denotes the nn-dimensional column vector with entries in CbC_b equal to 11 and 00 otherwise. Note that all entries of m‾\overline{m} corresponding to neurons in community CbC_b share the same value λζb\lambda\zeta_b, that is, m‾\overline{m} is constant within each community. The vector ζ\zeta can be explicitly computed from the k×kk\times k matrix WW defined in (16) whose entry WabW_{ab} corresponds to the mean cumulative effect that community CbC_b exerts on community CaC_a.

The proof of Theorem 3.1 can be divided into three steps. In the first step, we prove that for all admissible realizations of the environment, we can write

E(Θ,H)[X0]=(In−AT)−1λ1n,\mathbb{E}_{(\Theta,H)}[X_0]=(I_n-A^{\mathsf{T}})^{-1}\lambda\mathbf{1}_n,

where InI_n is the nn-dimensional identify matrix and 1n\mathbf{1}_n is the nn-dimensional column vector with all entries equal to 11. This is a consequence of Proposition 2.1 stated in § 2.2.

The second step consists in quantifying the error of approximating the random vector E(Θ,H)[X0]\mathbb{E}_{(\Theta,H)}[X_0] by the deterministic vector (In−E(n)[AT])−1λ1n(I_n-\mathbb{E}^{(n)}[A^{\mathsf{T}}])^{-1}\lambda\mathbf{1}_n, where E(n)\mathbb{E}^{(n)} denotes the expectation taken with respect to the randomness of the environment (Θ,H)(\Theta,H). This follows from Lemma A.1 of Appendix § 5 and Hoeffding’s inequality. The final step is then to show that (In−E(n)[AT])−1λ1n=m‾(I_n-\mathbb{E}^{(n)}[A^{\mathsf{T}}])^{-1}\lambda\mathbf{1}_n=\overline{m} inherits the block structure of the environment (Θ,H)(\Theta,H). Lemma D.2 of Appendix § 8 establishes the main result needed in the proof of this final step.

As a simple consequence of Theorem 3.1, we show that the vector of firing variances, defined as

v(Θ,H)(X0):=(E(Θ,H)[X1,0](1−E(Θ,H)[X1,0]),…,E(Θ,H)[Xn,0](1−E(Θ,H)[Xn,0]))T,v_{(\Theta,H)}(X_0) := \left(\mathbb{E}_{(\Theta,H)}[X_{1,0}]\left(1-\mathbb{E}_{(\Theta,H)}[X_{1,0}]\right),\ldots,\mathbb{E}_{(\Theta,H)}[X_{n,0}]\left(1-\mathbb{E}_{(\Theta,H)}[X_{n,0}]\right)\right)^{\mathsf{T}},

satisfies with high probability

∥v(Θ,H)(X0)−v‾∥∞≤Klog⁡(n)n,\left\|v_{(\Theta,H)}(X_0)-\overline{v}\right\|_{\infty}\leq K\sqrt{\frac{\log(n)}{n}},

where the nn-dimensional vector

v‾:=(m‾1(1−m‾1),…,m‾n(1−m‾n))T(3)\overline{v}:=\left(\overline{m}_1(1-\overline{m}_1),\ldots,\overline{m}_n(1-\overline{m}_n)\right)^{\mathsf{T}} \tag*{(3)}

is also constant within each community (see Corollary 3.1).

Next, we describe how the community structure of the environment emerges in the lagged firing covariances Σ(Θ,H)(t)\Sigma^{(t)}_{(\Theta,H)} as n→∞n\to\infty. We start describing what happens for the simultaneous firing covariance, that is for t=0t=0. In this case, since the diagonal terms of Σ(Θ,H)(0)\Sigma^{(0)}_{(\Theta,H)} are already known (they are given by the vector of firing variances v(Θ,H)(X0)v_{(\Theta,H)}(X_0)), we only need to focus on the behavior of its off-diagonal entries. The asymptotic behavior of these entries is described by means of a k×kk\times k matrix Ψ\Psi that is the unique solution of the Stein matrix equation

Ψ−WΨWT=WDWT,\Psi-W\Psi W^{\mathsf{T}}=WDW^{\mathsf{T}},

where DD is a diagonal matrix whose non-null entries are expressed in terms of the vector ζ\zeta defined in (17) and the parameters αb\alpha_b (see (20)). Specifically, introducing the n×nn\times n matrix

References

  1. [1][Abbe(2018)] Emmanuel Abbe. 2018. Community Detection and Stochastic Block Models: Recent Developments. Journal of Machine Learning Research 18, 177 (2018), 1–86. http://jmlr.org/papers/v18/16-480.html
  2. [2][Berthet et al.(2019)] Quentin Berthet, Philippe Rigollet, and Piyush Srivastava. 2019. Exact recovery in the Ising blockmodel. The Annals of Statistics 47, 4 (2019), 1805 – 1834. doi:10.1214/17-AOS1620DOI
  3. [3][Bunea et al.(2020)] Florentina Bunea, Christophe Giraud, Xi Luo, Martin Royer, and Nicolas Verzelen. 2020. Model assisted variable clustering: Minimax-optimal recovery and algorithms. The Annals of Statistics 48, 1 (2020), 111 – 137. doi:10.1214/18-AOS1794DOI
  4. [4][Chevallier et al.(2026)] Julien Chevallier, Eva Löcherbach, and Guilherme Ost. 2026. Inferring the dependence graph density of binary graphical models in high dimension. The Annals of Statistics 54, 2 (2026), 861 – 881. doi:10.1214/25-AOS2592DOI
  5. [5][Chevallier and Ost(2026)] Julien Chevallier and Guilherme Ost. 2026+. Community detection for binary graphical models in high dimension. To appear in Annals of Statistics (2026+).
  6. [6][Duarte et al.(2019)] Aline Duarte, Antonio Galves, Eva Löcherbach, and Guilherme Ost. 2019. Estimating the interaction graph of stochastic neural dynamics. Bernoulli 25, 1 (2019), 771–792.DOI
  7. [7][Endo et al.(2021)] Daisuke Endo, Ryota Kobayashi, Ramon Bartolo, Bruno B Averbeck, Yasuko Sugase-Miyamoto, Kazuko Hayashi, Kenji Kawano, Barry J Richmond, and Shigeru Shinomoto. 2021. A convolutional neural network for estimating synaptic connectivity from spike trains. Scientific Reports 11, 1 (2021), 12087.DOI
  8. [8][Horn and Johnson(1991)] Roger A. Horn and Charles R. Johnson. 1991. Topics in Matrix Analysis. Cambridge University Press, Cambridge.DOI
  9. [9][Kobayashi et al.(2019)] Ryota Kobayashi, Shuhei Kurita, Anno Kurth, Katsunori Kitano, Kenji Mizuseki, Markus Diesmann, Barry J Richmond, and Shigeru Shinomoto. 2019. Reconstructing neuronal circuitry from parallel spike trains. Nature Communications 10, 1 (2019), 4468.DOI
  10. [10][Linderman et al.(2016)] Scott Linderman, Ryan P Adams, and Jonathan W Pillow. 2016. Bayesian latent structure discovery from multi-neuron recordings. Advances in Neural Information Processing Systems (NeurIPS) 29 (2016).arxiv.org/abs/1610.08465
  11. [11][Ost and Reynaud-Bouret(2020)] Guilherme Ost and Patricia Reynaud-Bouret. 2020. Sparse space-time models: Concentration Inequalities and Lasso. Annales de l’Institut Henri Poincaré (B) Probabilité et Statistique 56, 4 (2020), 2377–2405.arxiv.org/abs/1807.07615
  12. [13][Peixoto(2019)] Tiago P Peixoto. 2019. Network reconstruction and community detection from dynamics. Physical Review Letters 123, 12 (2019), 128301.DOI
  13. [14][Pernice et al.(2011)] Volker Pernice, Benjamin Staude, Stefano Cardanobile, and Stefan Rotter. 2011. How Structure Determines Correlations in Neuronal Networks. PLoS Computational Biology 7, 5 (2011), e1002059. doi:10.1371/journal.pcbi.1002059DOI

Paper details

vlnc:W_qm4wy6t8
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 9

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents