Hierarchical Inference with Bayesian Neural Networks: An Application to Strong Gravitational Lensing
Abstract
In the past few years, approximate Bayesian Neural Networks (BNNs) have demonstrated the ability to produce statistically consistent posteriors on a wide range of inference problems at unprecedented speed and scale. However, any disconnect between training sets and the distribution of real-world objects can introduce bias when BNNs are applied to data. This is a common challenge in astrophysics and cosmology, where the unknown distribution of objects in our universe is often the science goal. In this work, we incorporate BNNs with flexible posterior parameterizations into a hierarchical inference framework that allows for the reconstruction of population hyperparameters and removes the bias introduced by the training distribution. We focus on the challenge of producing posterior PDFs for strong gravitational lens mass model parameters given Hubble Space Telescope–quality single-filter, lens-subtracted, synthetic imaging data. We show that the posterior PDFs are sufficiently accurate (statistically consistent with the truth) across a wide variety of power-law elliptical lens mass distributions. We then apply our approach to test data sets whose lens parameters are drawn from distributions that are drastically different from the training set. We show that our hierarchical inference framework mitigates the bias introduced by an unrepresentative training set's interim prior. Simultaneously, wemore »
- Authors:
-
- Stanford Univ., CA (United States); SLAC National Accelerator Lab., Menlo Park, CA (United States)
- Publication Date:
- Research Org.:
- SLAC National Accelerator Lab., Menlo Park, CA (United States)
- Sponsoring Org.:
- USDOE Office of Science (SC); National Science Foundation (NSF)
- Contributing Org.:
- LSST Dark Energy Science Collaboration
- OSTI Identifier:
- 1772052
- Alternate Identifier(s):
- OSTI ID: 1772050
- Report Number(s):
- SLAC-PUB-17592
Journal ID: ISSN 0004-637X; TRN: US2206944
- Grant/Contract Number:
- AC02-76SF00515; AC02-05CH11231; AST-1716527
- Resource Type:
- Accepted Manuscript
- Journal Name:
- The Astrophysical Journal
- Additional Journal Information:
- Journal Volume: 909; Journal Issue: 2; Journal ID: ISSN 0004-637X
- Publisher:
- Institute of Physics (IOP)
- Country of Publication:
- United States
- Language:
- English
- Subject:
- 79 ASTRONOMY AND ASTROPHYSICS; Strong gravitational lensing; Cosmology; Computational methods; Convolutional neural networks; Hierarchical models
Citation Formats
Wagner-Carena, Sebastian, Park, Ji Won, Birrer, Simon, Marshall, Philip J., Roodman, Aaron, and Wechsler, Risa H. Hierarchical Inference with Bayesian Neural Networks: An Application to Strong Gravitational Lensing. United States: N. p., 2021.
Web. doi:10.3847/1538-4357/abdf59.
Wagner-Carena, Sebastian, Park, Ji Won, Birrer, Simon, Marshall, Philip J., Roodman, Aaron, & Wechsler, Risa H. Hierarchical Inference with Bayesian Neural Networks: An Application to Strong Gravitational Lensing. United States. https://doi.org/10.3847/1538-4357/abdf59
Wagner-Carena, Sebastian, Park, Ji Won, Birrer, Simon, Marshall, Philip J., Roodman, Aaron, and Wechsler, Risa H. Wed .
"Hierarchical Inference with Bayesian Neural Networks: An Application to Strong Gravitational Lensing". United States. https://doi.org/10.3847/1538-4357/abdf59. https://www.osti.gov/servlets/purl/1772052.
@article{osti_1772052,
title = {Hierarchical Inference with Bayesian Neural Networks: An Application to Strong Gravitational Lensing},
author = {Wagner-Carena, Sebastian and Park, Ji Won and Birrer, Simon and Marshall, Philip J. and Roodman, Aaron and Wechsler, Risa H.},
abstractNote = {In the past few years, approximate Bayesian Neural Networks (BNNs) have demonstrated the ability to produce statistically consistent posteriors on a wide range of inference problems at unprecedented speed and scale. However, any disconnect between training sets and the distribution of real-world objects can introduce bias when BNNs are applied to data. This is a common challenge in astrophysics and cosmology, where the unknown distribution of objects in our universe is often the science goal. In this work, we incorporate BNNs with flexible posterior parameterizations into a hierarchical inference framework that allows for the reconstruction of population hyperparameters and removes the bias introduced by the training distribution. We focus on the challenge of producing posterior PDFs for strong gravitational lens mass model parameters given Hubble Space Telescope–quality single-filter, lens-subtracted, synthetic imaging data. We show that the posterior PDFs are sufficiently accurate (statistically consistent with the truth) across a wide variety of power-law elliptical lens mass distributions. We then apply our approach to test data sets whose lens parameters are drawn from distributions that are drastically different from the training set. We show that our hierarchical inference framework mitigates the bias introduced by an unrepresentative training set's interim prior. Simultaneously, we can precisely reconstruct the population hyperparameters governing our test distributions. Our full pipeline, from training to hierarchical inference on thousands of lenses, can be run in a day. The framework presented here will allow us to efficiently exploit the full constraining power of future ground- and space-based surveys (https://github.com/swagnercarena/ovejero).},
doi = {10.3847/1538-4357/abdf59},
journal = {The Astrophysical Journal},
number = 2,
volume = 909,
place = {United States},
year = {Wed Mar 17 00:00:00 EDT 2021},
month = {Wed Mar 17 00:00:00 EDT 2021}
}
Works referenced in this record:
Ensemble samplers with affine invariance
journal, January 2010
- Goodman, Jonathan; Weare, Jonathan
- Communications in Applied Mathematics and Computational Science, Vol. 5, Issue 1
Uncertainties in Parameters Estimated with Neural Networks: Application to Strong Gravitational Lensing
journal, November 2017
- Perreault Levasseur, Laurence; Hezaveh, Yashar D.; Wechsler, Risa H.
- The Astrophysical Journal, Vol. 850, Issue 1
Constraining the Reionization History using Bayesian Normalizing Flows
journal, September 2020
- Hortúa, Héctor J.; Malagò, Luigi; Volpi, Riccardo
- Machine Learning: Science and Technology, Vol. 1, Issue 3
Improving time-delay cosmography with spatially resolved kinematics
journal, September 2017
- Shajib, Anowar J.; Treu, Tommaso; Agnello, Adriano
- Monthly Notices of the Royal Astronomical Society, Vol. 473, Issue 1
Strong Lensing by Galaxies
journal, August 2010
- Treu, Tommaso
- Annual Review of Astronomy and Astrophysics, Vol. 48, Issue 1
Fast Calculation of a Family of Elliptical Gravitational Lens Models
journal, August 1998
- Barkana, Rennan
- The Astrophysical Journal, Vol. 502, Issue 2
The Sloan Lens ACS Survey. III. The Structure and Formation of Early‐Type Galaxies and Their Evolution since z ≈ 1
journal, October 2006
- Koopmans, Leon V. E.; Treu, Tommaso; Bolton, Adam S.
- The Astrophysical Journal, Vol. 649, Issue 2
The Sloan Lens ACS Survey. VII. Elliptical Galaxy Scaling Laws from Direct Observational Mass Measurements
journal, September 2008
- Bolton, Adam S.; Treu, Tommaso; Koopmans, Léon V. E.
- The Astrophysical Journal, Vol. 684, Issue 1
Parameter estimation for the cosmic microwave background with Bayesian neural networks
journal, November 2020
- Hortúa, Héctor J.; Volpi, Riccardo; Marinelli, Dimitri
- Physical Review D, Vol. 102, Issue 10
Calibration of the Tip of the Red Giant Branch
journal, March 2020
- Freedman, Wendy L.; Madore, Barry F.; Hoyt, Taylor
- The Astrophysical Journal, Vol. 891, Issue 1
Two Accurate Time-Delay Distances from Strong Lensing: Implications for Cosmology
journal, March 2013
- Suyu, S. H.; Auger, M. W.; Hilbert, S.
- The Astrophysical Journal, Vol. 766, Issue 2
Mining for Dark Matter Substructure: Inferring Subhalo Population Properties from Strong Lenses with Machine Learning
journal, November 2019
- Brehmer, Johann; Mishra-Sharma, Siddharth; Hermans, Joeri
- The Astrophysical Journal, Vol. 886, Issue 1
Direct detection of dark matter substructure in strong lens images with convolutional neural networks
journal, January 2020
- Diaz Rivero, Ana; Dvorkin, Cora
- Physical Review D, Vol. 101, Issue 2
Gravitationally lensed quasars and supernovae in future wide-field optical imaging surveys: Lensed quasars and supernovae
journal, April 2010
- Oguri, Masamune; Marshall, Philip J.
- Monthly Notices of the Royal Astronomical Society
lenstronomy: Multi-purpose gravitational lens modelling software package
journal, December 2018
- Birrer, Simon; Amara, Adam
- Physics of the Dark Universe, Vol. 22
STRIDES: a 3.9 per cent measurement of the Hubble constant from the strong lens system DES J0408−5354
journal, March 2020
- Shajib, A. J.; Birrer, S.; Treu, T.
- Monthly Notices of the Royal Astronomical Society, Vol. 494, Issue 4
TDCOSMO: IV. Hierarchical time-delay cosmography – joint inference of the Hubble constant and galaxy density profiles
journal, November 2020
- Birrer, S.; Shajib, A. J.; Galan, A.
- Astronomy & Astrophysics, Vol. 643
The sl2s Galaxy-Scale lens Sample. iii. lens Models, Surface Photometry, and Stellar Masses for the Final Sample
journal, October 2013
- Sonnenfeld, Alessandro; Gavazzi, Raphaël; Suyu, Sherry H.
- The Astrophysical Journal, Vol. 777, Issue 2
Inferring the Eccentricity Distribution
journal, December 2010
- Hogg, David W.; Myers, Adam D.; Bovy, Jo
- The Astrophysical Journal, Vol. 725, Issue 2
Planck 2018 results: VI. Cosmological parameters
journal, September 2020
- Aghanim, N.; Akrami, Y.; Ashdown, M.
- Astronomy & Astrophysics, Vol. 641, A6
HOLISMOKES: IV. Efficient mass modeling of strong lenses through deep learning
journal, February 2021
- Schuldt, S.; Suyu, S. H.; Meinhardt, T.
- Astronomy & Astrophysics, Vol. 646
Differentiable strong lensing: uniting gravity and neural nets through differentiable probabilistic programming
journal, May 2020
- Chianese, Marco; Coogan, Adam; Hofma, Paul
- Monthly Notices of the Royal Astronomical Society, Vol. 496, Issue 1
Massive Dark Matter Halos and Evolution of Early‐Type Galaxies to z ≈ 1
journal, August 2004
- Treu, Tommaso; Koopmans, Leon V. E.
- The Astrophysical Journal, Vol. 611, Issue 2
emcee : The MCMC Hammer
journal, March 2013
- Foreman-Mackey, Daniel; Hogg, David W.; Lang, Dustin
- Publications of the Astronomical Society of the Pacific, Vol. 125, Issue 925
H0LiCOW – XIII. A 2.4 per cent measurement of H0 from lensed quasars: 5.3σ tension between early- and late-Universe probes
journal, September 2019
- Wong, Kenneth C.; Suyu, Sherry H.; Chen, Geoff C-F
- Monthly Notices of the Royal Astronomical Society, Vol. 498, Issue 1
Leveraging uncertainty information from deep neural networks for disease detection
journal, December 2017
- Leibig, Christian; Allken, Vaneeda; Ayhan, Murat Seçkin
- Scientific Reports, Vol. 7, Issue 1
Exoplanet Population Inference and the Abundance of Earth Analogs from Noisy, Incomplete Catalogs
journal, October 2014
- Foreman-Mackey, Daniel; Hogg, David W.; Morton, Timothy D.
- The Astrophysical Journal, Vol. 795, Issue 1
Large Magellanic Cloud Cepheid Standards Provide a 1% Foundation for the Determination of the Hubble Constant and Stronger Evidence for Physics beyond ΛCDM
journal, May 2019
- Riess, Adam G.; Casertano, Stefano; Yuan, Wenlong
- The Astrophysical Journal, Vol. 876, Issue 1
The Carnegie-Chicago Hubble Program. VIII. An Independent Determination of the Hubble Constant Based on the Tip of the Red Giant Branch
journal, August 2019
- Freedman, Wendy L.; Madore, Barry F.; Hatt, Dylan
- The Astrophysical Journal, Vol. 882, Issue 1
The sl2s Galaxy-Scale lens Sample. v. dark Matter Halos and Stellar imf of Massive Early-Type Galaxies out to Redshift 0.8
journal, February 2015
- Sonnenfeld, Alessandro; Treu, Tommaso; Marshall, Philip J.
- The Astrophysical Journal, Vol. 800, Issue 2
The Sloan Lens ACS Survey. I. A Large Spectroscopically Selected Sample of Massive Early‐Type Lens Galaxies
journal, February 2006
- Bolton, Adam S.; Burles, Scott; Koopmans, Leon V. E.
- The Astrophysical Journal, Vol. 638, Issue 2
The Population of Galaxy–Galaxy Strong Lenses in Forthcoming Optical Imaging Surveys
journal, September 2015
- Collett, Thomas E.
- The Astrophysical Journal, Vol. 811, Issue 1
Tensions between the early and late Universe
journal, September 2019
- Verde, Licia; Treu, Tommaso; Riess, Adam G.
- Nature Astronomy, Vol. 3, Issue 10
Cosmic Shear with Einstein Rings
journal, January 2018
- Birrer, Simon; Refregier, Alexandre; Amara, Adam
- The Astrophysical Journal, Vol. 852, Issue 1
Is every strong lens model unhappy in its own way? Uniform modelling of a sample of 13 quadruply+ imaged quasars
journal, December 2018
- Shajib, A. J.; Birrer, S.; Treu, T.
- Monthly Notices of the Royal Astronomical Society, Vol. 483, Issue 4
Cosmological Applications of Gravitational Lensing
journal, September 1992
- Blandford, R. D.; Narayan, R.
- Annual Review of Astronomy and Astrophysics, Vol. 30, Issue 1