Systematic fluctuation expansion for neural network activity equations

Population rate or activity equations are the foundation of a common approach to modeling for neural networks. These equations provide mean field dynamics for the firing rate or activity of neurons within a network given some connectivity. The shortcoming of these equations is that they take into account only the average firing rate while leaving out higher order statistics like correlations between firing. A stochastic theory of neural networks which includes statistics at all orders was recently formulated. We describe how this theory yields a systematic extension to population rate equations by introducing equations for correlations and appropriate coupling terms. Each level of the approximation yields closed equations, i.e. they depend only upon the mean and specific correlations of interest, without an {\it ad hoc} criterion for doing so. We show in an example of an all-to-all connected network how our system of generalized activity equations captures phenomena missed by the mean fieldrate equations alone.

💡 Research Summary

The paper addresses a fundamental limitation of traditional population‑rate (mean‑field) models of neural networks: they describe only the average firing rate while ignoring higher‑order statistics such as pairwise correlations. Building on a recently formulated stochastic theory that retains statistics of all orders, the authors develop a systematic fluctuation expansion that extends the mean‑field framework in a principled way. Starting from the master equation for the full stochastic network, they treat the inverse system size (1/N) as a small parameter and expand the dynamics order by order. The zeroth‑order term recovers the classic mean‑field rate equation. The first‑order correction adds a linear term to the mean rate, but the most important contribution appears at second order, where the dynamics of the mean rate become coupled to the dynamics of the two‑point correlation (covariance) function. Crucially, the resulting set of equations is closed: the second‑order system involves only the mean and the chosen correlation, without any ad‑hoc truncation rule.

Mathematically, the expansion reveals that the non‑linearity of the neuronal transfer function (e.g., a sigmoid) generates coupling terms that feed the covariance back into the mean‑field dynamics. These terms become dominant near critical points or under strong external noise, producing phenomena that the plain mean‑field model cannot capture, such as oscillations, multistability, and abrupt transitions. To demonstrate the utility of the approach, the authors apply the second‑order equations to an all‑to‑all connected network. While the mean‑field model predicts only a single stable fixed point, the extended system exhibits limit‑cycle oscillations and coexistence of multiple stable states, in agreement with full stochastic simulations.

The authors also discuss computational advantages. Tracking only a few macroscopic variables (the mean rate and a limited set of correlations) dramatically reduces dimensionality compared to simulating the full Markov chain, yet retains essential dynamical features. Numerical experiments show that going beyond second or third order yields diminishing returns for realistic network sizes, suggesting that low‑order expansions already provide a good balance between accuracy and efficiency.

In summary, the work presents a rigorous, systematic method for incorporating fluctuations and correlations into neural population models. By deriving closed, hierarchy‑truncated equations without arbitrary assumptions, it bridges the gap between simple mean‑field descriptions and full stochastic network simulations. This framework has broad implications for theoretical neuroscience, the analysis of learning dynamics in artificial neural networks, and the modeling of brain‑machine interfaces or pathological brain states where correlations play a pivotal role.