Saved in:
Bibliographic Details
Main Authors: Trinh, Trung, Heinonen, Markus, Acerbi, Luigi, Kaski, Samuel
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2306.02775
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914702939914240
author Trinh, Trung
Heinonen, Markus
Acerbi, Luigi
Kaski, Samuel
author_facet Trinh, Trung
Heinonen, Markus
Acerbi, Luigi
Kaski, Samuel
contents Deep Ensembles (DEs) demonstrate improved accuracy, calibration and robustness to perturbations over single neural networks partly due to their functional diversity. Particle-based variational inference (ParVI) methods enhance diversity by formalizing a repulsion term based on a network similarity kernel. However, weight-space repulsion is inefficient due to over-parameterization, while direct function-space repulsion has been found to produce little improvement over DEs. To sidestep these difficulties, we propose First-order Repulsive Deep Ensemble (FoRDE), an ensemble learning method based on ParVI, which performs repulsion in the space of first-order input gradients. As input gradients uniquely characterize a function up to translation and are much smaller in dimension than the weights, this method guarantees that ensemble members are functionally different. Intuitively, diversifying the input gradients encourages each network to learn different features, which is expected to improve the robustness of an ensemble. Experiments on image classification datasets and transfer learning tasks show that FoRDE significantly outperforms the gold-standard DEs and other ensemble methods in accuracy and calibration under covariate shift due to input perturbations.
format Preprint
id arxiv_https___arxiv_org_abs_2306_02775
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Input-gradient space particle inference for neural network ensembles
Trinh, Trung
Heinonen, Markus
Acerbi, Luigi
Kaski, Samuel
Machine Learning
Deep Ensembles (DEs) demonstrate improved accuracy, calibration and robustness to perturbations over single neural networks partly due to their functional diversity. Particle-based variational inference (ParVI) methods enhance diversity by formalizing a repulsion term based on a network similarity kernel. However, weight-space repulsion is inefficient due to over-parameterization, while direct function-space repulsion has been found to produce little improvement over DEs. To sidestep these difficulties, we propose First-order Repulsive Deep Ensemble (FoRDE), an ensemble learning method based on ParVI, which performs repulsion in the space of first-order input gradients. As input gradients uniquely characterize a function up to translation and are much smaller in dimension than the weights, this method guarantees that ensemble members are functionally different. Intuitively, diversifying the input gradients encourages each network to learn different features, which is expected to improve the robustness of an ensemble. Experiments on image classification datasets and transfer learning tasks show that FoRDE significantly outperforms the gold-standard DEs and other ensemble methods in accuracy and calibration under covariate shift due to input perturbations.
title Input-gradient space particle inference for neural network ensembles
topic Machine Learning
url https://arxiv.org/abs/2306.02775