Wavelet compression for sampling deep Gaussian processes
Abstract
Deep Gaussian processes (DGPs) use iterative sampling to create non-stationary random fields. This is notoriously expensive because the covariance at each non-initial layer depends on the realised field from the layer below, so it changes with every sample. Moreover, standard fast sampling techniques like circulant embedding or matrix compression fail since the underlying kernel usually lacks the necessary regularity and structure. We develop and analyse a wavelet compression method for convolution Matérn DGPs. At each layer the covariance matrix is compressed using wavelets and a Lanczos algorithm computes the random sample. For specific parameter ranges involving the smoothness of the Matérn kernel as well as the smoothness of the wavelets, we show that the accumulative compression and sampling errors through all layers decay exponentially with increasing truncation parameter. We obtain a practical DGP sampling algorithm by choosing a truncation that gives log-linear covariance storage and matrix-vector costs, enabling DGP sampling at nearly linear computational cost.
Related
- Canonical Layer Inference for Deep Gaussian Processes via Symmetry-Breaking Augmentation
- Continuous-depth Deep Gaussian Processes
- Quasihelix properties of selected Volterra Gaussian processes
- Gaussian approximation for the integrated intensity of counting processes: application to Hawkes processes
- UPPER TAIL PROBABILITY ESTIMATES OF SUPREMA OF GAUSSIAN PROCESSES BY MATRICES, A NUMBER THEORETIC EXAMPLE
BibTeX
@misc{broukal2026wavelet,
title = {{Wavelet compression for sampling deep Gaussian processes}},
author = {Oskar Broukal and Michael Feischl and Conor Osborne},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.20266},
eprint = {2610.20266},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.20266},
}
RIS
TY - UNPB
TI - Wavelet compression for sampling deep Gaussian processes
AU - Oskar Broukal
AU - Michael Feischl
AU - Conor Osborne
PY - 2026
DA - 2026/10/14
PB - arXiv
DO - 10.48550/arxiv.2610.20266
UR - https://arxiv.org/abs/2610.20266
AB - Deep Gaussian processes (DGPs) use iterative sampling to create non-stationary random fields. This is notoriously expensive because the covariance at each non-initial layer depends on the realised field from the layer below, so it changes with every sample. Moreover, standard fast sampling techniques like circulant embedding or matrix compression fail since the underlying kernel usually lacks the necessary regularity and structure. We develop and analyse a wavelet compression method for convolution Matérn DGPs. At each layer the covariance matrix is compressed using wavelets and a Lanczos algorithm computes the random sample. For specific parameter ranges involving the smoothness of the Matérn kernel as well as the smoothness of the wavelets, we show that the accumulative compression and sampling errors through all layers decay exponentially with increasing truncation parameter. We obtain a practical DGP sampling algorithm by choosing a truncation that gives log-linear covariance storage and matrix-vector costs, enabling DGP sampling at nearly linear computational cost.
ER -