Collective Noise Filtering in Complex Networks
Tingyu Zhao, István A. Kovács · 2026
Read Collective Noise Filtering in Complex Networks by Tingyu Zhao, István A. Kovács (2026) 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-2601-21299.txt.
Introduction
Tingyu Zhao and István A. Kovács work on cleaning noise out of network data. Networks are a standard way to picture complex systems, from cells to societies. Real network data, though, often carry noise, bias, and missing links, and the weights on links can be wrong.
Bad weights can spoil every later study built on the network. The authors present a way to filter noise out of the whole network at once. Cleaning the data first makes later claims about groups, hubs, and spread more reliable. The approach is meant for the messy data that real projects actually collect.
Catalog record
| CREATOR | Tingyu Zhao, István A. Kovács |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 38 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-21299
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Collective Noise Filtering in Complex Networks Tingyu Zhao, István A. Kovács Abstract Complex networks are powerful representations of complex systems across scales and domains, and the field is experiencing unprecedented growth in data availability. However, real-world network data often suffer from noise, biases, and missing data in the edge weights, which undermine the reliability of downstream network analyses. Standard noise filtering approaches, whether treating individual edges one-by-one or assuming a uniform global noise level, are suboptimal, because in reality both signal and noise can be heterogeneous and correlated across multiple edges. As a solution, we introduce the Network Wiener Filter, a principled method for collective edge-level noise filtering that leverages both network topology and noise characteristics, to reduce error in the observed edge weights and to infer missing edge weights. We demonstrate the broad practical efficacy of the Network Wiener Filter in two distinct settings, the genetic interaction network of the yeast S. cerevisiae and the Enron Corpus email network, noting compelling evidence of successful noise suppression in both applications. With the Network Wiener Filter, we advocate for a shift toward error-aware network science, one that embraces data imperfection as an inherent feature and learns to navigate it effectively. Ever since the early development of graph theory, mathematicians have expanded its toolkit far beyond elegant puzzles, building increasingly sophisticated theories for reasoning about connectivity and structure 14 . In recent decades, network science transformed these abstract ideas into the empirical world, using networks to represent complex systems across diverse disciplines, including, among others, social science 36 , biology 3 , transportation 5 , climate science 21 , finance and economics 1 . Analyzing these complex systems from a network perspective has enabled researchers to uncover hidden patterns, capture emergent properties, and generate unique insights 4 . While the network science community has benefited from a rapid expansion in the availability of network data, it is frequently overlooked that such data can be imperfect. Real-world networks are often subjected to various kinds of errors such as noise, biases, and incomplete observation 41 ; 33 ; 58 ; 19 . Yet it remains common in the literature to omit any quantitative assessment or reporting of these errors 46 . Moreover, most network analysis pipelines do not consider potential data errors and treat empirical networks as if they were exact graphs, in the spirit of graph theory. Ignoring the issue of data reliability in this manner can compromise the robustness of downstream network-based inference, leading to flawed conclusions and decisions. In response, a growing body of work has sought to mitigate edge noise in network data by exploiting network topological information, including link prediction methods 37 ; 18 ; 64 ; 38 ; 7 ; 39 ; 63 ; 32 ; 60 , diffusion-based methods 22 ; 59 ; 62 , generative model-based methods 15 ; 28 ; 45 ; 44 ; 50 ; 49 , and singular value shrinkage methods 16 ; 9 ; 52 ; 43 ; 24 ; 29 ; 25 ; 55 . Each method category relies on different, often strong, network structural assumptions and is tailored to different use cases (SI Table [placeholder]). However, none of these approaches is adequate for the realistic scenario of having heterogeneous and correlated noise. Either noise level is not explicitly incorporated, as in link prediction and diffusion-based methods, or noise is assumed to be largely homogeneous and independent across edges, as in generative model-based and singular value shrinkage methods. In the case of heterogeneous, or even correlated, noise, practitioners often resort to local thresholding strategies at the level of individual edges, without harnessing the rich information in the network architecture 19 . This, in turn, can limit the amount of signal that can be uncovered from the noisy data. In this work, we propose the Network Wiener Filter (NetWF) as a principled method for collective noise filtering in complex networks. Grounded in the classical theory of Wiener filtering, the NetWF takes into account network topology through an intuitive edge-similarity measure, which is then combined with explicit edge-level noise statistics in a synergistic way. Importantly, this formulation naturally accommodates heterogeneous or correlated noise across edges, as well as the special case of homogeneous and independent noise, rendering it fully general with respect to noise structure. At the same time, the NetWF applies seamlessly to directed or undirected, binary, weighted or even signed networks. As illustrated on our real-world applications, our efficient implementation using the conjugate gradient algorithm enables the NetWF to scale to large empirical networks. From generalized Wiener filter to the NetWF In classical signal processing, the Wiener filter is a foundational framework designed to recover the underlying true signal from noisy observations by minimizing the expected mean square error (MSE) between the estimate and the (unknown) true signal 47 . Using the statistical properties of both the signal and the noise, it seeks an optimal balance between noise suppression and signal preservation. As a result, Wiener filtering has been widely adopted for denoising tasks such as in time series (Figure 1 A) and image processing (Figure 1 B). Generalized Wiener filter Here, we start with a generalized Wiener filter formalism 51 and extend it to network data. Imagine that we observe a demeaned data vector a=u+n\in\mathbb{R}^{k} , where u denotes the true signal and n the additive noise. For data types that are not inherently one-dimensional, a can be understood as a vectorized representation, such as a vectorized image or a vectorized network adjacency matrix, a=\mathrm{vec}(A) , where A_{ij} stands for the interaction weight between nodes i and j in the network. The goal is then to identify a “Wiener operator” G^{\mathrm{W}} that, when acting on a , gives an estimate of the signal \hat{u}=G^{\mathrm{W}}a that minimizes the mean square error (MSE): \displaystyle G^{\mathrm{W}}=\,\arg\min_{G}\,\mathbb{E}\left\lVert Ga-u\right\rVert^{2}_{2}, (1) where the expectation \mathbb{E} is taken over the joint realizations of u and n , and \lVert\cdot\rVert_{2} is the L^{2} norm. The optimization problem of Eq. 1 admits a closed-form analytical solution, see SI Text [placeholder]. In the case when u and n are drawn from independent distributions, the solution is given by \displaystyle G^{\mathrm{W}}=C_{u}(C_{u}+C_{n})^{-1}, (2) in terms of the covariance matrices C_{u}=\mathbb{E}[uu^{T}] and C_{n}=\mathbb{E}[nn^{T}] , both with dimensionality \mathbb{R}^{k\times k} . Figure 1: Illustration of Wiener filtering for noise reduction across data types, and why networks are different. Starting from (A) a time series, (B) an image, and (C) a bipartite toy network, we add independent and identically distributed Gaussian white noise to obtain noisy observations. We then apply Wiener filtering to estimate the underlying signal from noisy observations using the known noise statistics (more details in SI Text [placeholder] and SI Figure [placeholder]). In time series and images, the classical Wiener filter exploits data-independent temporal or spatial proximity to capture signal correlations, with the intuition that nearby samples tend to be similar. In contrast, network data is inherently high-dimensional and requires a data-dependent adaptive treatment. The NetWF bridges this gap by using a network-informed edge-similarity measure that captures correlations globally across edges, as discussed in the main text. In the toy network, our NetWF reveals the original structure of two communities. Applying the Wiener filter hinges on having access to the two covariance