Interpolation of Multifeature Time-Vertex Signals via Joint Spectral ARMA Model
Year: 2026
Type: conference-paper
Abstract: Graph signal processing provides effective techniques for the analysis of data structures residing on irregular topologies. Existing time-vertex signal processing approaches in the literature generally focus on single-attribute signals. However, many applications require working with data types where multiple attributes (e.g., temperature, humidity, wind speed) are measured simultaneously at each node. In this study, a novel Multi-feature Joint Spectral ARMA (MF-JS-ARMA) model based on the Joint Wide Sense Stationarity assumption is proposed for modeling multi-attribute time-vertex signals and estimating missing observations in partially observed data. The proposed method simultaneously learns the auto-spectral and cross-spectral densities of attribute pairs within the framework of a joint optimization problem. The developed model has been applied to the LMMSE-based missing data estimation problem on meteorological data, and it has been shown to improve estimation performance compared to reference methods in the literature.
Authors: Enfal Çabuk, Elif Vural
Institution: Middle East Technical University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7202193203