Federated automatic latent variable selection in multi-output Gaussian processes

Jingyi Gao

doi:10.1016/j.patcog.2025.112410

Federated automatic latent variable selection in multi-output Gaussian processes

2025-09-08

0

PRE

AI

DOI：10.1016/j.patcog.2025.112410

Original

Original request for help

Save

Abstract

En 中文

• We build a linear model of coregionalization that (i) does not require raw data sharing for estimating cross-unit covariances and (ii) can automatically infer the necessary number of latent functions to represent shared patterns across units. • We propose a federated learning framework that estimates the parameters of our proposed model and makes personalized predictions for each unit, without requiring centralized computation or the disclosure of units’ local data. • We propose an efficient learning approach for new units entering the system, based on the important latent functions identified through our automatic variable selection. • We demonstrate the effectiveness of our proposed model in two real-world applications using reliability engineering and climate data.

Federated automatic latent variable selection in multi-output Gaussian processes

Abstract

Journal

Organization