Distributional gaussian process layers for outlier detection in image
segmentation
segmentation
File(s)
Working paper
Author(s)
Popescu, Sebastian G
Sharp, David J
Cole, James H
Kamnitsas, Konstantinos
Glocker, Ben
Type
Conference Paper
Abstract
We propose a parameter efficient Bayesian layer for hierarchical
convolutional Gaussian Processes that incorporates Gaussian Processes operating
in Wasserstein-2 space to reliably propagate uncertainty. This directly
replaces convolving Gaussian Processes with a distance-preserving affine
operator on distributions. Our experiments on brain tissue-segmentation show
that the resulting architecture approaches the performance of well-established
deterministic segmentation algorithms (U-Net), which has never been achieved
with previous hierarchical Gaussian Processes. Moreover, by applying the same
segmentation model to out-of-distribution data (i.e., images with pathology
such as brain tumors), we show that our uncertainty estimates result in
out-of-distribution detection that outperforms the capabilities of previous
Bayesian networks and reconstruction-based approaches that learn normative
distributions.
convolutional Gaussian Processes that incorporates Gaussian Processes operating
in Wasserstein-2 space to reliably propagate uncertainty. This directly
replaces convolving Gaussian Processes with a distance-preserving affine
operator on distributions. Our experiments on brain tissue-segmentation show
that the resulting architecture approaches the performance of well-established
deterministic segmentation algorithms (U-Net), which has never been achieved
with previous hierarchical Gaussian Processes. Moreover, by applying the same
segmentation model to out-of-distribution data (i.e., images with pathology
such as brain tumors), we show that our uncertainty estimates result in
out-of-distribution detection that outperforms the capabilities of previous
Bayesian networks and reconstruction-based approaches that learn normative
distributions.
Date Issued
2021-04-28
Date Acceptance
2021-02-12
Citation
2021, 12729
Publisher
arXiv
Volume
12729
Copyright Statement
© 2021 The Author(s). This version is published under CC BY license.
License URL
Sponsor
Innovate UK
Identifier
http://arxiv.org/abs/2104.13756v1
Grant Number
104691
Source
Information Processing in Medical Imaging (IPMI) 2021
Subjects
stat.ML
stat.ML
cs.LG
Publication Status
Published
Start Date
2021-06-28
Coverage Spatial
Virtual