A prediction rigidity formalism for low-cost uncertainties in trained neural networks

Quantifying the uncertainty of regression models is essential to ensure their reliability, particularly since their application often extends beyond their training domain. Based on the solution of a constrained optimization problem, this work proposes ‘prediction rigidities’ as a formalism to obtain...

Full description

Saved in:
Bibliographic Details
Main Authors: Filippo Bigi, Sanggyu Chong, Michele Ceriotti, Federico Grasselli
Format: Article
Language:English
Published: IOP Publishing 2024-01-01
Series:Machine Learning: Science and Technology
Subjects:
Online Access:https://doi.org/10.1088/2632-2153/ad805f
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850210725225562112
author Filippo Bigi
Sanggyu Chong
Michele Ceriotti
Federico Grasselli
author_facet Filippo Bigi
Sanggyu Chong
Michele Ceriotti
Federico Grasselli
author_sort Filippo Bigi
collection DOAJ
description Quantifying the uncertainty of regression models is essential to ensure their reliability, particularly since their application often extends beyond their training domain. Based on the solution of a constrained optimization problem, this work proposes ‘prediction rigidities’ as a formalism to obtain uncertainties of arbitrary pre-trained regressors. A clear connection between the suggested framework and Bayesian inference is established, and a last-layer approximation is developed and rigorously justified to enable the application of the method to neural networks. This extension affords cheap uncertainties without any modification to the neural network itself or its training procedure. The effectiveness of this approach is shown for a wide range of regression tasks, ranging from simple toy models to applications in chemistry and meteorology.
format Article
id doaj-art-761be39486c44b488e70c67939201fda
institution OA Journals
issn 2632-2153
language English
publishDate 2024-01-01
publisher IOP Publishing
record_format Article
series Machine Learning: Science and Technology
spelling doaj-art-761be39486c44b488e70c67939201fda2025-08-20T02:09:43ZengIOP PublishingMachine Learning: Science and Technology2632-21532024-01-015404501810.1088/2632-2153/ad805fA prediction rigidity formalism for low-cost uncertainties in trained neural networksFilippo Bigi0https://orcid.org/0000-0002-9338-7317Sanggyu Chong1https://orcid.org/0000-0002-6948-1602Michele Ceriotti2https://orcid.org/0000-0003-2571-2832Federico Grasselli3https://orcid.org/0000-0003-4284-0094Laboratory of Computational Science and Modeling, Institut des Matériaux , École Polytechnique Fédérale de Lausanne, 1015 Lausanne, SwitzerlandLaboratory of Computational Science and Modeling, Institut des Matériaux , École Polytechnique Fédérale de Lausanne, 1015 Lausanne, SwitzerlandLaboratory of Computational Science and Modeling, Institut des Matériaux , École Polytechnique Fédérale de Lausanne, 1015 Lausanne, SwitzerlandLaboratory of Computational Science and Modeling, Institut des Matériaux , École Polytechnique Fédérale de Lausanne, 1015 Lausanne, Switzerland; Department of Physics, Informatics and Mathematics, University of Modena and Reggio Emilia , via Giuseppe Campi 213/a, Modena 41125, ItalyQuantifying the uncertainty of regression models is essential to ensure their reliability, particularly since their application often extends beyond their training domain. Based on the solution of a constrained optimization problem, this work proposes ‘prediction rigidities’ as a formalism to obtain uncertainties of arbitrary pre-trained regressors. A clear connection between the suggested framework and Bayesian inference is established, and a last-layer approximation is developed and rigorously justified to enable the application of the method to neural networks. This extension affords cheap uncertainties without any modification to the neural network itself or its training procedure. The effectiveness of this approach is shown for a wide range of regression tasks, ranging from simple toy models to applications in chemistry and meteorology.https://doi.org/10.1088/2632-2153/ad805fpredictionsrigiditylow-cost uncertaintiesneural networkregressionpre-trained
spellingShingle Filippo Bigi
Sanggyu Chong
Michele Ceriotti
Federico Grasselli
A prediction rigidity formalism for low-cost uncertainties in trained neural networks
Machine Learning: Science and Technology
predictions
rigidity
low-cost uncertainties
neural network
regression
pre-trained
title A prediction rigidity formalism for low-cost uncertainties in trained neural networks
title_full A prediction rigidity formalism for low-cost uncertainties in trained neural networks
title_fullStr A prediction rigidity formalism for low-cost uncertainties in trained neural networks
title_full_unstemmed A prediction rigidity formalism for low-cost uncertainties in trained neural networks
title_short A prediction rigidity formalism for low-cost uncertainties in trained neural networks
title_sort prediction rigidity formalism for low cost uncertainties in trained neural networks
topic predictions
rigidity
low-cost uncertainties
neural network
regression
pre-trained
url https://doi.org/10.1088/2632-2153/ad805f
work_keys_str_mv AT filippobigi apredictionrigidityformalismforlowcostuncertaintiesintrainedneuralnetworks
AT sanggyuchong apredictionrigidityformalismforlowcostuncertaintiesintrainedneuralnetworks
AT micheleceriotti apredictionrigidityformalismforlowcostuncertaintiesintrainedneuralnetworks
AT federicograsselli apredictionrigidityformalismforlowcostuncertaintiesintrainedneuralnetworks
AT filippobigi predictionrigidityformalismforlowcostuncertaintiesintrainedneuralnetworks
AT sanggyuchong predictionrigidityformalismforlowcostuncertaintiesintrainedneuralnetworks
AT micheleceriotti predictionrigidityformalismforlowcostuncertaintiesintrainedneuralnetworks
AT federicograsselli predictionrigidityformalismforlowcostuncertaintiesintrainedneuralnetworks