Communities in the Kuramoto Model: Dynamics and Detection via Path Signatures

Tâm J Nguyên 1 , Darrick Lee 2 and Bernadette J Stolz 1,3 · 2025

Read Communities in the Kuramoto Model: Dynamics and Detection via Path Signatures by Tâm J Nguyên 1 , Darrick Lee 2 and Bernadette J Stolz 1,3 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-17546.txt.

Introduction

Nguyên, Lee, and Stolz recover hidden communities from oscillator chatter using path signatures. Their work appeared in twenty twenty-five. Brains and other networks reveal their wiring only through activity, and plain correlation often misses the pattern.

Path signatures compress a wiggly time series into features that keep its shape and timing. From them, a lead matrix exposes which components run ahead of others. The test bed couples Kuramoto oscillators on a block-structured graph. Mean-field math and Gaussian approximations predict the lead matrix in each time regime. The resulting algorithm restored the planted communities exactly in several trials. A noisy variant and real cortical recordings from Neuropixels probes stretched the method further.

Catalog record

CREATORTâm J Nguyên 1 , Darrick Lee 2 and Bernadette J Stolz 1,3
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE127 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-17546

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Communities in the Kuramoto Model: Dynamics and Detection via Path Signatures Tâm J Nguyên 1 , Darrick Lee 2 and Bernadette J Stolz 1,3 Abstract . The behavior of multivariate dynamical processes is often governed by underlying structural connections that relate the components of the system. For example, brain activity which is often measured via time series is determined by an underlying structural graph, where nodes represent neurons or brain regions and edges represent cortical connectivity. Existing methods for inferring structural connections from observed dynamics, such as correlation-based or spectral techniques, may fail to fully capture complex relationships in high-dimensional time series in an interpretable way. Here, we propose the use of path signatures—a mathematical framework that encodes geometric and temporal properties of continuous paths—to address this problem. Path signatures provide a reparametrization-invariant characterization of dynamical data and, in particular, can be used to compute the lead matrix which reveals lead-lag phenomena. We showcase our approach on time series from coupled oscillators in the Kuramoto model defined on a stochastic block model graph, termed the Kuramoto stochastic block model (KSBM). Using mean-field theory and Gaussian approximations, we analytically derive reduced models of KSBM dynamics in different temporal regimes and theoretically characterize the lead matrix in these settings. Leveraging these insights, we propose a novel signature-based community detection algorithm, achieving exact recovery of structural communities from observed time series in multiple KSBM instances. Our results demonstrate that path signatures provide a novel perspective on analyzing complex neural data and other high-dimensional systems, explicitly exploiting temporal functional relationships to infer underlying structure. 1. Introduction Figure 1. Community estimation in the Kuramoto stochastic block model (A) The Kuramoto stochastic block model (KSBM) is a version of the Kuramoto model with underlying coupling between oscillators defined by a stochastic block model graph. Time series \Theta arising from the KSBM can be split into different temporal regimes with different macro-scale properties. (B) We investigate high-dimensional time series using path signatures. From the path signatures, we derive lead matrices L(\Theta) and L(\sin(\Theta)) which capture lead and lag behaviour. When computed over the (full) time series produced by our KSBM, i.e. ignoring the different time regimes shown in (A), high noise prevents identification of communities in L(\Theta) . When splitting the time series into different time regimes, we find clearer block patterns in L(\Theta) for the transient regime and in L(\sin(\Theta)) for the transient and steady state regimes. (C) We propose a novel clustering algorithm based on path signatures and lead matrices that estimates communities in the KSBM’s underlying stochastic block model graph from time series split into different time regimes. Our algorithm is based on maximizing a block clustering metric g(D|G)/n of the distance matrix D over community assignment G with n communities. Many dynamical processes in our physical world are rooted in structured interactions between entities of interest. This underlying structure is often not accessible, and we are often limited to observing resulting dynamical processes through time series from experiments. In many research fields such as neuroscience, estimating structural interactions which give rise to functional activity of a system is of great interest [ 52 , 44 , 45 , 50 , 51 , 29 ] . 1.1. Motivation: Neuroscience Cognitive function is determined by anatomical connections between neurons or brain regions. These underlying connections can be interpreted as structural networks in which the neurons or brain regions are nodes and their connections represent edges. Identifying the underlying structural networks that give rise to specific cognitive functions is one of the key questions in experimental neuroscience [ 42 ] . However, due to experimental and ethical constraints, direct observation of these structural connections is often impossible. Instead, we typically study neural activity, which reflects dynamics arising in connected neurons or brain regions. Linking this observed neural activity to the underlying structural network is a complex challenge. Brain activity arises from the simultaneous occurence of multiple neural processes involving many neurons; the resulting data is thus often high-dimensional and noisy. Biologically, brain activity is driven by overlapping circuits of connected neurons, which fire in response to inputs from other neurons. In a neural circuit, the axon connection between a presynaptic and a postsynaptic neuron induces time synchrony in the responses of the connected neurons, i.e. the firing of the presynaptic neuron will typically lead to a time-delayed firing in the postsynaptic neuron. This neural activity can be represented in a lower-dimensional latent space, known as the neural manifold [ 27 ] , which enables the study of phenomena such as synchrony [ 21 ] . In this latent space, the time-delayed firing induced by synaptic connections reflects a form of synchronization between the activity of connected neurons with a temporal offset. The degree of synchronization in neural activity is referred to as functional connectivity [ 41 , 47 , 11 , 10 , 40 ] , which can be measured from coupled time series data, such as fMRI or EEG, using a variety of methods. A common approach is to calculate the pairwise cross correlation between time series. This functional connectivity can then be used to construct a functional network, where nodes represent neurons or brain regions, and edges are weighted according to the degree of functional connectivity. Such networks can be analyzed using techniques from graph theory or network science 1 1 1 Note that one can also study coupled time series using other tools, e.g. [ 52 , 39 , 49 ] . [ 41 , 47 , 11 , 10 , 40 ] . However, because neurons often participate in multiple circuits, functional connectivity alone is typically insufficient to distinguish between different circuits and thus directly infer structural connections. The highly structured nature of brain circuits allows us to utilize knowledge about neuronal organization, such as their partition into densely-connected communities, to facilitate the inference of structural connections. Let us consider a motivating example that demonstrates principles of neuronal organisation. Cortical processes of perception and sensation are active in most background brain activities. The neural activity generated by these cortical processes involves sensory cells first transducing their sensory inputs to the primary cortical areas followed by the synchronization of cortical columns. These columns consist of neurons whose activity continuously represents the perceived space (known as topographic mapping 2 2 2 Mapping is understood here in the biological sense: each neuron is activated by a specific part of the space of stimuli, called a receptive field. Formally, we could write the neural activity map as f:X\times Y\rightarrow\mathbb{R} , where X is some embedding of the cortex and Y the space of stimuli. Neighbourhoods of neurons are activated by neighbourhoods of receptive fields, hence this mapping of the stimuli space to the activity of neuron is continuous. in neuroscience [ 23 ] ). For example, in auditory perception the perceived space is the space of frequencies perceivable by a human. This space is represented by cortical columns in the primary auditory cortex [ 42 ] ; each column is activated by the presence of a specific frequency in the perceived sound. Those columns continuously represent lower to higher frequencies, preserving the structure of the frequency space. Due to this continuous representation, neurons in the same

Read the complete text