Abstract:Graph signal processing (GSP) is a prominent framework for analyzing signals on non-Euclidean domains. The graph Fourier transform (GFT) uses the combinatorial graph Laplacian matrix to reveal the spectral decomposition of signals in the graph frequency domain. However, a common challenge in applying GSP methods is that in many scenarios the underlying graph of a system is unknown. A solution in such cases is to construct the unobserved graph from available data, which is commonly referred to as graph or network inference. Although different graph inference methods exist, these are restricted to learning from either smooth graph signals or simple additive Gaussian noise. Other types of noisy data, such as discrete counts or binary digits, are rather common in real-world applications, yet are underexplored in graph inference. In this paper, we propose a versatile graph inference framework for learning from graph signals corrupted by exponential family noise. Our framework generalizes previous methods from continuous smooth graph signals to various data types. We propose an alternating algorithm that jointly estimates the graph Laplacian and the unobserved smooth representation from the noisy signals. We also extend our approach to a variational form to account for the inherent stochasticity of the latent smooth representation. Finally, since real-world graph signals are frequently non-independent and temporally correlated, we further adapt our original setting to a time-vertex formulation. We demonstrate on synthetic and real-world data that our new algorithms outperform competing Laplacian estimation methods that suffer from noise model mismatch.

Graph Laplacian Matrix Learning from Smooth Time-Vertex Signal

Time-varying graph learning from smooth and stationary graph signals with hidden nodes

Graph Laplacian Learning with Exponential Family Noise

Graph Learning for Latent-Variable Gaussian Graphical Models under Laplacian Constraints

Time-Varying Graph Signal Reconstruction

Learning Spatiotemporal Graphical Models From Incomplete Observations

Graph Learning for Spatiotemporal Signals with Long- and Short-Term Characterization

A Time-Vertex Signal Processing Framework

Graph signal reconstruction based on spatio-temporal features learning

Learning Laplacian Matrix for Smooth Signals on Graph

Time-Varying Graph Signal Recovery Using High-Order Smoothness and Adaptive Low-rankness

Graph Learning Based on Spatiotemporal Smoothness for Time-Varying Graph Signal

Learning Cartesian Product Graphs with Laplacian Constraints

Time-varying Signals Recovery via Graph Neural Networks

Graph-based Reconstruction of Time-Varying Spatial Signals

A Distributed Algorithm for Reconstructing Time-Varying Graph Signals

Joint Signal Recovery and Graph Learning from Incomplete Time-Series

Graph Learning from Data under Structural and Laplacian Constraints

Graph Learning for Spatiotemporal Dynamic Signal

Graph Learning from Gaussian and Stationary Graph Signals

Towards stationary time-vertex signal processing