Multiresolution Analysis and Statistical Thresholding on Dynamic Networks
Raphaël Romero, Tijl De Bie, Nick Heard, Alexander Modell · 2025
Read Multiresolution Analysis and Statistical Thresholding on Dynamic Networks by Raphaël Romero, Tijl De Bie, Nick Heard, Alexander Modell (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-2506-01208.txt.
Introduction
Raphaël Romero, Tijl De Bie, Nick Heard, and Alexander Modell tackle in 2025 the problem of spotting change in a network that never sits still. Slicing time into bins and comparing snapshots is the usual method, but bin width forces a trade: fine bins catch fast change yet rest on thin, noisy data.
Borrowing the logic of signal processing, the authors analyze the network at several time resolutions at once and use statistical thresholding to decide which changes are real. The approach matters most where anomalies hide at unknown speeds, like cyberattacks. Choosing the scale becomes part of the answer rather than a guess made beforehand.
Catalog record
| CREATOR | Raphaël Romero, Tijl De Bie, Nick Heard, Alexander Modell |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 60 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-01208
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Multiresolution Analysis and Statistical Thresholding on Dynamic Networks Raphaël Romero, Tijl De Bie, Nick Heard, Alexander Modell Abstract Detecting structural change in dynamic network data has wide-ranging applications. Existing approaches typically divide the data into time bins, extract network features within each bin, and then compare these features over time. This introduces an inherent tradeoff between temporal resolution and the statistical stability of the extracted features. Despite this tradeoff, reminiscent of time–frequency tradeoffs in signal processing, most methods rely on a fixed temporal resolution . Choosing an appropriate resolution parameter is typically difficult, and can be especially problematic in domains like cybersecurity, where anomalous behavior may emerge at multiple time scales. We address this challenge by proposing ANIE ( A daptive N etwork I ntensity E stimation), a multi-resolution framework designed to automatically identify the time scales at which network structure evolves, enabling the joint detection of both rapid and gradual changes. Modeling interactions as Poisson processes, our method proceeds in two steps: (1) estimating a low-dimensional subspace of node behavior, and (2) deriving a set of novel empirical affinity coefficients that quantify change in interaction intensity between latent factors and support statistical testing for structural change across time scales. We provide theoretical guarantees for subspace estimation and the asymptotic behavior of the affinity coefficients, enabling model-based change detection. Experiments on synthetic networks show that ANIE adapts to the appropriate time resolution, and is able to capture sharp structural changes while remaining robust to noise. Furthermore, applications to real-world data showcase the practical benefits of ANIE ’s multiresolution approach to detecting structural change over fixed resolution methods. An open-source implementation of the method is available at https://github.com/aida-ugent/anie . 1 Introduction Understanding dynamic networks, namely datasets taking the form of sequences of interaction events (u,v,t) between nodes u and v at timestamp t has wide-ranging applications in domains such as contact tracing [ 15 ] , cybersecurity [ 32 ] and urban mobility studies [ 2 , 20 ] . Despite this domain diversity, temporal networks commonly exhibit two fundamental types of structure: cross-sectional structure , where the network is seen as a graph evolving over time, and longitudinal structure , where the data at its finest resolution is best modeled as a collection of point processes [ 31 , 28 , 27 ] . At its core, change detection in such networks involves understanding how these two types of structure interact. However, doing so involves an inherent tradeoff. On one hand, identifying cross-sectional structure-such as communities-requires aggregating events over a sufficiently wide time window to achieve statistical stability. On the other hand, imposing a certain resolution of analysis may obscure transient events which occur at higher temporal resolutions. This mirrors the time–frequency tradeoff in signal processing, where narrow time windows reveal high-frequency details but miss low-frequency trends, and wide windows improve frequency resolution at the cost of temporal localization. In practice, the choice of an appropriate time resolution is a challenge which manifests in a variety of ways, such as selecting the number of timesteps at which to evaluate dynamic node embeddings [ 38 , 39 ] , or selecting a bandwidth in order to derive smooth temporal signals from the dynamic network [ 27 ] . Often this challenge is resolved by selecting a resolution which seems to correspond to some characteristic period or frequency derived a priori from domain knowledge [ 16 , 19 , 27 ] . However in applications such as cybersecurity [ 31 , 33 , 12 ] , where time-localization of anomalous event is essential, or more broadly community detection [ 40 ] , where node behaviors may align at different resolution levels, such an arbitrary choice of resolution is not satisfactory. To resolve this paradox, we highlight that cross-sectional structure in real-world networks typically manifests at several resolution levels simultaneously. For instance in social networks, community events of a few hours coexist with gradually evolving friendship structures (weeks to months). Similarly, in cybersecurity, malicious activity might include both rapid bursts of suspicious connections and slowly evolving patterns designed to evade detection [ 33 ] . On the other hand, in bike-sharing networks [ 10 ] , interaction patterns exhibit daily rhythms (commuting), weekly cycles (workday vs. weekend usage), and seasonal trends (weather effects). In this paper, we introduce ANIE ( A daptive N etwork I ntensity E stimation), a novel approach for detecting changes in dynamic networks across multiple temporal resolutions. Our approach takes inspiration in recent work in multiresolution analysis of point process [ 6 , 42 , 13 , 11 ] , and more generally wavelet analysis [ 26 ] , and adapts them to the network domain. Contributions Our contributions can be summarized as follows: In Section 3 , we formulate change detection as a statistical signal processing problem, where the goal is to recover edge-level temporal signals from noisy dynamic network observations. In Section 4 , we present a new statistical method for multi-resolution change detection, supported by theoretical guarantees. In Section 5 , we evaluate our method on both synthetic and real-world datasets, demonstrating that ANIE outperforms fixed-resolution approaches by effectively capturing changes at multiple time scales in dynamic networks. 2 Related Work The proposed work lies at the intersection of several fields, which we briefly overview below. Change Detection in Dynamic Networks The task of understanding the temporal evolution of dynamic network structure has been approached from various angles. One common approach is to view it as a dimensionality reduction task: constructing time-varying statistical summaries—such as node embeddings [ 38 , 27 ] —and tracking their evolution over time. Recent work in this direction includes dynamic extensions of spectral clustering [ 45 , 29 , 41 , 27 ] , latent space models [ 38 , 39 ] , and tensor factorization methods [ 24 , 36 , 46 , 1 , 16 , 43 ] , which represent temporal structure through evolving latent factors. A related line of work focuses specifically on detecting change points, often in an online setting, by comparing network summaries across time windows [ 19 , 12 , 47 ] . These methods typically rely on fixed time intervals, which assumes short-term stationarity. In contrast, our approach adaptively adjusts the time resolution based on the data—addressing a key limitation of existing methods. Wavelets and Point Process Intensity Estimation Wavelet analysis has been proposed as a principled approach to addressing the time–frequency tradeoff in signal processing [ 26 , 44 ] , and has proven effective in estimating the intensity of single point processes [ 8 , 14 , 13 , 11 , 42 , 48 , 22 ] , where key features of the intensity function often appear at multiple resolution levels. To our knowledge, our work is the to integrate these wavelet-based point process analysis with a low-rank decomposition of cross-sectional network structure. Functional Data Analysis Functional Data Analysis (FDA) has been widely used to analyze data with a continuous time dimension [ 37 ] , and has been extended to multivariate settings [ 17 ] . Recent work has also adapted FDA to point process observations [ 34 ] . This work is the first to explicitly apply similar techniques