A Residual Tree Gaussian Process Modeling Framework for High-Dimensional Data

Abstract

With the advance of measurement technologies and increasing computing power, large spatial data with heterogeneous structures are often collected over high-dimensional domains. Existing Gaussian process (GP) models and computational strategies are often inadequate for analyzing such datasets in multi-dimensional domains. To address these challenges, we develop a Bayesian residual tree GP methodology called ResTGP for large spatial data with potentially heterogeneous structures in multi-dimensional domains. The key idea is to decompose a Gaussian process at a cascade of resolutions along a dyadic tree through iteratively computing predictive and residual processes so that the residual process on each tree node, both interior and leaf, becomes sufficient for the finer-level dependency within that node. This allows characterization of the underlying covariance structure in a flexible, multi-scale manner while achieving divide-and-conquer on the data domain, which leads to computational efficiency. To allow efficient tree inference, we introduce a computational strategy for Bayesian inference based on recursive message passing, which scales linearly with the sample size given the tree. This paper also proves posterior consistency of the model for estimating continuous functions in a nonparametric regression framework. Extensive numerical examples and the storm surge application confirm the advantages of the proposed method.

The authors’ abstract, as arXiv lists it. arXiv’s listings, abstracts included, are free for anyone to reuse.

More in Methodology

• arXiv preprint · 3 October 2026

  Gradient-Free Sampling from Generative Models via Stochastic Bounded Extremum Seeking

  Alexander Scheinker
• arXiv preprint · 1 October 2026

  On skew-symmetric distributions and their use in Monte Carlo sampling algorithms: coordinate-free, Gibbs-style and manifold versions of the Barker proposal

  Minh Vu, Samuel Livingstone and Pantelis Samartsidis
• arXiv preprint · 3 October 2026

  Efficient Mass Matrix Estimation with Gaussian Cooling

  Jake Hofgard and Michael Lindsey
• arXiv preprint · 1 October 2026

  MCIR: A Feature Dependence-Aware Explainability Method with Reliability Guarantees

  Poushali Sengupta, Sabita Maharjan, Frank Eliassen and 2 others
• arXiv preprint · 3 October 2026

  Low-Rank and Structured Sparse Tensor Decomposition for Anomaly Detection in Multivariate Functional Data

  Mohammad N. Bisheh, Che-Yi Liao and Kamran Paynabar

All new papers in Methodology