ePrints@IIScePrints@IISc Home | About | Browse | Latest Additions | Advanced Search | Contact | Help

Product Graph Gaussian Processes for Multi-domain Data Imputation and Active Learning

Kadambari, SK and Chepuri, SP (2023) Product Graph Gaussian Processes for Multi-domain Data Imputation and Active Learning. In: 31st European Signal Processing Conference, EUSIPCO 2023, 4 - 8 September 2023, Helsinki, pp. 1619-1623.

[img] PDF
Eur_sig_pro_con_2023.pdf - Published Version
Restricted to Registered users only

Download (966kB) | Request a copy
Official URL: https://doi.org/10.23919/EUSIPCO58844.2023.1029002...


In this work, we consider the problem of imputing signals defined on the nodes of a product graph from the subset of observations. To this end, we focus on learning their predictive probability distribution function (PDF) based Gaussian processes. In particular, we propose a product graph Gaussian process model, which incorporates the product graph structure in the Gaussian process kernel via a product graph filter. When the observed graph signals are real-valued, the mean and variance of the predictive PDF can be computed in closed form. Further, the variance captures the model uncertainty, which we use for active learning to obtain subsequent observations. We demonstrate the efficacy of the proposed method on multi-domain data imputation and active learning tasks on synthetic and real-world datasets. © 2023 European Signal Processing Conference, EUSIPCO. All rights reserved.

Item Type: Conference Paper
Publication: European Signal Processing Conference
Publisher: European Signal Processing Conference, EUSIPCO
Additional Information: The copyright for this article belongs to European Signal Processing Conference, EUSIPCO.
Keywords: Distribution functions; Gaussian noise (electronic); Graph theory; Graphic methods; Learning systems; Supervised learning; Uncertainty analysis, Active Learning; Data imputation; Gaussian process models; Gaussian Processes; Graph; Multi-domains; Probability distribution functions; Product graph; Semi-supervised learning; Semi-supervised learning over graph, Gaussian distribution
Department/Centre: Division of Electrical Sciences > Electrical Communication Engineering
Date Deposited: 01 Mar 2024 05:35
Last Modified: 01 Mar 2024 05:35
URI: https://eprints.iisc.ac.in/id/eprint/83803

Actions (login required)

View Item View Item