Detecting gravitational lenses using machine learning: exploring interpretability and sensitivity to rare lensing configurations

Abstract

Forthcoming large imaging surveys such as Euclid and the Vera Rubin Observatory Legacy Survey of Space and Time are expected to find more than $10^5$ strong gravitational lens systems, including many rare and exotic populations such as compound lenses, but these $10^5$ systems will be interspersed among much larger catalogues of $\sim 10^9$ galaxies. This volume of data is too much for visual inspection by volunteers alone to be feasible and gravitational lenses will only appear in a small fraction of these data which could cause a large amount of false positives. Machine learning is the obvious alternative but the algorithms' internal workings are not obviously interpretable, so their selection functions are opaque and it is not clear whether they would select against important rare populations. We design, build, and train several Convolutional Neural Networks (CNNs) to identify strong gravitational lenses using VIS, Y, J, and H bands of simulated data, with F1 scores between 0.83 and 0.91 on 100,000 test set images. We demonstrate for the first time that such CNNs do not select against compound lenses, obtaining recall scores as high as 76\% for compound arcs and 52\% for double rings. We verify this performance using Hubble Space Telescope (HST) and Hyper Suprime-Cam (HSC) data of all known compound lens systems. Finally, we explore for the first time the interpretability of these CNNs using Deep Dream, Guided Grad-CAM, and by exploring the kernels of the convolutional layers, to illuminate why CNNs succeed in compound lens selection.

arXiv commentMNRAS accepted. 16 pages, 17 figures, The code used in this paper is publicly available at github.com/JoshWilde/LensFindery-McLensFinderFace

References in corpus (53)

• scikit-image: Image processing in Python
• Deep Learning using Rectified Linear Units (ReLU)
• Understanding Neural Networks Through Deep Visualization
• H0LiCOW XIII. A 2.4% measurement of $H_0$ from lensed quasars: $5.3\sigma$ tension between early and late-Universe probes
• The Sloan Lens ACS Survey. V. The Full ACS Strong-Lens Sample
• The Detection of a Population of Submillimeter-Bright, Strongly-Lensed Galaxies
• The population of galaxy-galaxy strong lenses in forthcoming optical imaging surveys
• Lenstronomy: multi-purpose gravitational lens modelling software package
• The Sloan Lens ACS Survey. IX. Colors, Lensing and Stellar Masses of Early-type Galaxies
• Fast Automated Analysis of Strong Gravitational Lenses with Convolutional Neural Networks
• ALMA Imaging and Gravitational Lens Models of South Pole Telescope-Selected Dusty, Star-Forming Galaxies at High Redshifts
• The Sloan Lens ACS Survey. VI: Discovery and analysis of a double Einstein ring
• CMU DeepLens: Deep Learning For Automatic Image-based Galaxy-Galaxy Strong Lens Finding
• The Strong Gravitational Lens Finding Challenge
• Source-position transformation -- an approximate invariance in strong gravitational lensing
• First catalog of strong lens candidates in the COSMOS field
• The Herschel-ATLAS: a sample of 500μm-selected lensed galaxies over 600 square degrees
• Cosmological Constraints from the double source plane lens SDSSJ0946+1006
• Finding strong lenses in CFHTLS using convolutional neural networks
• Space Warps: I. Crowd-sourcing the Discovery of Gravitational Lenses
• Space Warps II. New Gravitational Lens Candidates from the CFHTLS Discovered through Citizen Science
• The Sloan Lens ACS Survey. XIII. Discovery of 40 New Galaxy-Scale Strong Lenses
• Finding high-redshift strong lenses in DES using convolutional neural networks
• The LSST Dark Energy Science Collaboration (DESC) Science Requirements Document
• Morpheus: A Deep Learning Framework For Pixel-Level Analysis of Astronomical Image Data
• Over-constrained Gravitational Lens Models and the Hubble Constant
• Identifying Strong Lenses with Unsupervised Machine Learning using Convolutional Autoencoder
• Recurrent Neural Networks to Correct Satellite Image Classification Maps
• Gravitational lenses and lens candidates identified from the COSMOS field
• Cosmology with the Large Synoptic Survey Telescope: an Overview
• Survey of Gravitationally-lensed Objects in HSC Imaging (SuGOHI). VI. Crowdsourced lens finding with Space Warps
• Constraining the dark energy equation of state with double source plane strong lenses
• High-quality strong lens candidates in the final Kilo Degree survey footprint
• Using Convolutional Neural Networks to identify Gravitational Lenses in Astronomical images
• The Sloan Lens ACS Survey. XI. Beyond Hubble resolution: size, luminosity and stellar mass of compact lensed galaxies at intermediate redshift
• NOEMA redshift measurements of bright Herschel galaxies
• AstroVaDEr: Astronomical Variational Deep Embedder for Unsupervised Morphological Classification of Galaxies and Synthetic Image Generation
• Testing Convolutional Neural Networks for finding strong gravitational lenses in KiDS
• KiDS-SQuaD II: Machine learning selection of bright extragalactic objects to search for new gravitationally lensed quasars
• The Red Radio Ring: a gravitationally lensed hyperluminous infrared radio galaxy at z=2.553 discovered through citizen science
• Deep Learning for Strong Lensing Search: Tests of the Convolutional Neural Networks and New Candidates from KiDS DR3
• Fanaroff-Riley classification of radio galaxies using group-equivariant convolutional neural networks
• The mass profile of early-type galaxies in overdense environments: the case of the double source plane gravitational lens SL2SJ02176-0513
• The inner dark matter distribution of the Cosmic Horseshoe (J1148+1930) with gravitational lensing and dynamics
• Support Vector Machine classification of strong gravitational lenses
• Powerful t-SNE technique leading to clear separation of type-2 AGN and HII galaxies in BPT diagrams
• A Spectroscopically Confirmed Double Source Plane Lens System in the Hyper Suprime-Cam Subaru Strategic Program
• Euclid: Nonparametric point spread function field recovery through interpolation on a graph Laplacian
• Finding bright $z\ge 6.6$ Lyman-$\alpha$ emitters with lensing: prospects for Euclid
• Super-resolving Herschel imaging: a proof of concept using Deep Neural Networks
• The Atacama Cosmology Telescope: CO(J = 3 - 2) mapping and lens modeling of an ACT-selected dusty star-forming galaxy
• Hubble Constant from LSST Strong lens time delays with microlensing systematics
• Strong Lensing considerations for the LSST observing strategy

Cited by in corpus (15)

• Astronomia ex machina: a history, primer, and outlook on neural networks in astronomy
• Sensitivity of strong lensing observations to dark matter substructure: a case study with Euclid
• Streamlined Lensed Quasar Identification in Multiband Images via Ensemble Networks
• HOLISMOKES -- XI. Evaluation of supervised neural networks for strong-lens searches in ground-based imaging surveys
• TEGLIE: Transformer encoders as strong gravitational lens finders in KiDS
• Developing a Victorious Strategy to the Second Strong Gravitational Lensing Data Challenge
• When Spectral Modeling Meets Convolutional Networks: A Method for Discovering Reionization-era Lensed Quasars in Multi-band Imaging Data
• Anisotropic strong lensing as a probe of dark matter self-interactions
• Cosmological Inference with Cosmic Voids and Neural Network Emulators
• Measuring the Substructure Mass Power Spectrum of 23 SLACS Strong Galaxy-Galaxy Lenses with Convolutional Neural Networks
• HOLISMOKES XV. Search for strong gravitational lenses combining ground-based and space-based imaging
• Transformers as Strong Lens Detectors- From Simulation to Surveys
• 200,000+ Deep Learning-inferred Periods of Stellar Variability from the All-Sky Automated Survey for Supernovae
• Assessing Galaxy Rotation Kinematics: Insights from Convolutional Neural Networks on Velocity Variations
• Predicting large scale cosmological structure evolution with generative adversarial network-based autoencoders