Robustness of Neural Ratio and Posterior Estimators to Distributional Shifts for Population-level Dark Matter Analysis in Strong Gravitational Lensing

We investigate the robustness of neural ratio estimators (NREs) and sequential neural posterior estimators (SNPEs) to distributional shifts in the context of measuring the abundance of dark matter subhalos using strong gravitational lensing data. While these data-driven inference frameworks can be a...

Full description

Saved in:
Bibliographic Details
Main Authors: Andreas Filipp, Yashar Hezaveh, Laurence Perreault-Levasseur
Format: Article
Language:English
Published: IOP Publishing 2025-01-01
Series:The Astrophysical Journal
Subjects:
Online Access:https://doi.org/10.3847/1538-4357/adee20
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849233389403504640
author Andreas Filipp
Yashar Hezaveh
Laurence Perreault-Levasseur
author_facet Andreas Filipp
Yashar Hezaveh
Laurence Perreault-Levasseur
author_sort Andreas Filipp
collection DOAJ
description We investigate the robustness of neural ratio estimators (NREs) and sequential neural posterior estimators (SNPEs) to distributional shifts in the context of measuring the abundance of dark matter subhalos using strong gravitational lensing data. While these data-driven inference frameworks can be accurate on test data from the same distribution as the training sets, in real applications, it is expected that simulated training data and true observational data will differ in their distributions. We explore the behavior of a trained NRE and trained SNPEs to estimate the population-level parameters of dark matter subhalos from a large sample of images of strongly lensed galaxies with test data presenting distributional shifts within and beyond the bounds of the training distribution in the nuisance parameters (e.g., the background source morphology). While our results show that NREs and SNPEs perform well when tested perfectly in distribution, they exhibit significant biases that often lead to not recovering the ground truth in the 3 σ interval when confronted with slight deviations from the examples seen in the training distribution. This indicates the necessity for caution when applying NREs and SNPEs to real astrophysical data, where high-dimensional underlying distributions are not perfectly known.
format Article
id doaj-art-41bce4e343334a54b8faf1e5cadd42ae
institution Kabale University
issn 1538-4357
language English
publishDate 2025-01-01
publisher IOP Publishing
record_format Article
series The Astrophysical Journal
spelling doaj-art-41bce4e343334a54b8faf1e5cadd42ae2025-08-20T06:01:48ZengIOP PublishingThe Astrophysical Journal1538-43572025-01-01989222610.3847/1538-4357/adee20Robustness of Neural Ratio and Posterior Estimators to Distributional Shifts for Population-level Dark Matter Analysis in Strong Gravitational LensingAndreas Filipp0https://orcid.org/0000-0003-4701-3469Yashar Hezaveh1https://orcid.org/0000-0002-8669-5733Laurence Perreault-Levasseur2https://orcid.org/0000-0003-3544-3939Ciela Institute , Montreal Institute for Astrophysics and Machine Learning, Montreal QC, Canada ; andreas.filipp@umontreal.ca; Mila—Quebec AI Institute , Montreal QC, Canada; Department of Physics, University of Montréal , Montréal QC, CanadaCiela Institute , Montreal Institute for Astrophysics and Machine Learning, Montreal QC, Canada ; andreas.filipp@umontreal.ca; Mila—Quebec AI Institute , Montreal QC, Canada; Department of Physics, University of Montréal , Montréal QC, Canada; Center for Computational Astrophysics, Flatiron Institute , NY, USA; Perimeter Institute for Theoretical Physics , Waterloo, Canada; Trottier Space Institute, McGill University , Montréal, CanadaCiela Institute , Montreal Institute for Astrophysics and Machine Learning, Montreal QC, Canada ; andreas.filipp@umontreal.ca; Mila—Quebec AI Institute , Montreal QC, Canada; Department of Physics, University of Montréal , Montréal QC, Canada; Center for Computational Astrophysics, Flatiron Institute , NY, USA; Perimeter Institute for Theoretical Physics , Waterloo, Canada; Trottier Space Institute, McGill University , Montréal, CanadaWe investigate the robustness of neural ratio estimators (NREs) and sequential neural posterior estimators (SNPEs) to distributional shifts in the context of measuring the abundance of dark matter subhalos using strong gravitational lensing data. While these data-driven inference frameworks can be accurate on test data from the same distribution as the training sets, in real applications, it is expected that simulated training data and true observational data will differ in their distributions. We explore the behavior of a trained NRE and trained SNPEs to estimate the population-level parameters of dark matter subhalos from a large sample of images of strongly lensed galaxies with test data presenting distributional shifts within and beyond the bounds of the training distribution in the nuisance parameters (e.g., the background source morphology). While our results show that NREs and SNPEs perform well when tested perfectly in distribution, they exhibit significant biases that often lead to not recovering the ground truth in the 3 σ interval when confronted with slight deviations from the examples seen in the training distribution. This indicates the necessity for caution when applying NREs and SNPEs to real astrophysical data, where high-dimensional underlying distributions are not perfectly known.https://doi.org/10.3847/1538-4357/adee20Strong gravitational lensingDark matterNeural networksGravitational lensingGalaxy dark matter halos
spellingShingle Andreas Filipp
Yashar Hezaveh
Laurence Perreault-Levasseur
Robustness of Neural Ratio and Posterior Estimators to Distributional Shifts for Population-level Dark Matter Analysis in Strong Gravitational Lensing
The Astrophysical Journal
Strong gravitational lensing
Dark matter
Neural networks
Gravitational lensing
Galaxy dark matter halos
title Robustness of Neural Ratio and Posterior Estimators to Distributional Shifts for Population-level Dark Matter Analysis in Strong Gravitational Lensing
title_full Robustness of Neural Ratio and Posterior Estimators to Distributional Shifts for Population-level Dark Matter Analysis in Strong Gravitational Lensing
title_fullStr Robustness of Neural Ratio and Posterior Estimators to Distributional Shifts for Population-level Dark Matter Analysis in Strong Gravitational Lensing
title_full_unstemmed Robustness of Neural Ratio and Posterior Estimators to Distributional Shifts for Population-level Dark Matter Analysis in Strong Gravitational Lensing
title_short Robustness of Neural Ratio and Posterior Estimators to Distributional Shifts for Population-level Dark Matter Analysis in Strong Gravitational Lensing
title_sort robustness of neural ratio and posterior estimators to distributional shifts for population level dark matter analysis in strong gravitational lensing
topic Strong gravitational lensing
Dark matter
Neural networks
Gravitational lensing
Galaxy dark matter halos
url https://doi.org/10.3847/1538-4357/adee20
work_keys_str_mv AT andreasfilipp robustnessofneuralratioandposteriorestimatorstodistributionalshiftsforpopulationleveldarkmatteranalysisinstronggravitationallensing
AT yasharhezaveh robustnessofneuralratioandposteriorestimatorstodistributionalshiftsforpopulationleveldarkmatteranalysisinstronggravitationallensing
AT laurenceperreaultlevasseur robustnessofneuralratioandposteriorestimatorstodistributionalshiftsforpopulationleveldarkmatteranalysisinstronggravitationallensing