arXiv subjects

T. Mitchell Roddenberry

Publications and source records attributed to T. Mitchell Roddenberry.

25 records · Page 2Linked to original sources

arXiv2020

Network topology change-point detection from graph signals with prior spectral signatures

We consider the problem of sequential graph topology change-point detection from graph signals. We assume that signals on the nodes of the graph are regularized by the underlying graph structure via a graph filtering model, which we then leverage to distill the graph topology change-point detection problem to a subspace detection problem. We demonstrate how prior information on the spectral signature of the post-change graph can be incorporated to implicitly denoise the observed sequential data, thus leading to a natural CUSUM-based algorithm for change-point detection. Numerical experiments illustrate the performance of our proposed approach, particularly underscoring the benefits of (potentially noisy) prior information.

stat.ML↗

arXiv2020

Network Topology Inference with Graphon Spectral Penalties

We consider the problem of inferring the unobserved edges of a graph from data supported on its nodes. In line with existing approaches, we propose a convex program for recovering a graph Laplacian that is approximately diagonalizable by a set of eigenvectors obtained from the second-order moment of the observed data. Unlike existing work, we incorporate prior knowledge about the distribution from where the underlying graph was drawn. In particular, we consider the case where the graph was drawn from a graphon model, and we supplement our convex optimization problem with a provably-valid regularizer on the spectrum of the graph to be recovered. We present the cases where the graphon model is assumed to be known and the more practical setting where the relevant features of the model are inferred from auxiliary network observations. Numerical experiments on synthetic and real-world data illustrate the advantage of leveraging the proposed graphon prior, even when the prior is imperfect.

eess.SP↗

arXiv2020

Blind Inference of Eigenvector Centrality Rankings

We consider the problem of estimating a network's eigenvector centrality only from data on the nodes, with no information about network topology. Leveraging the versatility of graph filters to model network processes, data supported on the nodes is modeled as a graph signal obtained via the output of a graph filter applied to white noise. We seek to simplify the downstream task of centrality ranking by bypassing network topology inference methods and, instead, inferring the centrality structure of the graph directly from the graph signals. To this end, we propose two simple algorithms for ranking a set of nodes connected by an unobserved set of edges. We derive asymptotic and non-asymptotic guarantees for these algorithms, revealing key features that determine the complexity of the task at hand. Finally, we illustrate the behavior of the proposed algorithms on synthetic and real-world datasets.

cs.SI↗

arXiv2020

Blind Estimation of Eigenvector Centrality from Graph Signals: Beyond Low-pass Filtering

This paper characterizes the difficulty of estimating a network's eigenvector centrality only from data on the nodes, i.e., with no information about the topology of the network. We model this nodal data as graph signals generated by passing white noise through generic (not necessarily low-pass) graph filters. Leveraging the spectral properties of graph filters, we estimate the eigenvectors of the adjacency matrix of the underlying network. To this end, a simple selection algorithm is proposed, which chooses the correct eigenvector of the signal covariance matrix with minimal assumptions on the underlying graph filter. We then present a theoretical characterization of the asymptotic and non-asymptotic performance of this algorithm, thus providing a sample complexity bound for the centrality estimation and revealing key elements driving this complexity. Finally, we illustrate the developed insights through a set of numerical experiments on different random graph models.

cs.SI↗

arXiv2020

Exact Blind Community Detection from Signals on Multiple Graphs

Networks and data supported on graphs have become ubiquitous in the sciences and engineering. This paper studies the 'blind' community detection problem, where we seek to infer the community structure of a graph model given the observation of independent graph signals on a set of nodes whose connections are unknown. We model each observation as filtered white noise, where the underlying network structure varies with every observation. These varying network structures are modeled as independent realizations of a latent planted partition model (PPM), justifying our assumption of a constant underlying community structure over all observations. Under certain conditions on the graph filter and PPM parameters, we propose algorithms for determining (i) the number of latent communities and (ii) the associated partitions of the PPM. We then prove statistical guarantees in the asymptotic and non-asymptotic sampling cases. Numerical experiments on real and synthetic data demonstrate the efficacy of our algorithms.

cs.SI↗

arXiv2019

HodgeNet: Graph Neural Networks for Edge Data

Networks and network processes have emerged as powerful tools for modeling social interactions, disease propagation, and a variety of additional dynamics driven by relational structures. Recently, neural networks have been generalized to process data on graphs, thus being able to learn from the aforementioned network processes achieving cutting-edge performance in traditional tasks such as node classification and link prediction. However, these methods have all been formulated in a way suited only to data on the nodes of a graph. The application of these techniques to data supported on the edges of a graph, namely flow signals, has not been explored in detail. To bridge this gap, we propose the use of the so-called Hodge Laplacian combined with graph neural network architectures for the analysis of flow data. Specifically, we apply two graph neural network architectures to solve the problems of flow interpolation and source localization.

eess.SP↗

arXiv2019

Blind Inference of Centrality Rankings from Graph Signals

We study the blind centrality ranking problem, where our goal is to infer the eigenvector centrality ranking of nodes solely from nodal observations, i.e., without information about the topology of the network. We formalize these nodal observations as graph signals and model them as the outputs of a network process on the underlying (unobserved) network. A simple spectral algorithm is proposed to estimate the leading eigenvector of the associated adjacency matrix, thus serving as a proxy for the centrality ranking. A finite rate performance analysis of the algorithm is provided, where we find a lower bound on the number of graph signals needed to correctly rank (with high probability) two nodes of interest. We then specialize our general analysis for the particular case of dense \ER graphs, where existing graph-theoretical results can be leveraged. Finally, we illustrate the proposed algorithm via numerical experiments in synthetic and real-world networks, making special emphasis on how the network features influence the performance.

cs.SI↗