CMU DeepLens: deep learning for automatic image-based galaxy–galaxy strong lens finding

Abstract

Galaxy-scale strong gravitational lensing is not only a valuable probe of the dark matter distribution of massive galaxies, but can also provide valuable cosmological constraints, either by studying the population of strong lenses or by measuring time delays in lensed quasars. Due to the rarity of galaxy-scale strongly lensed systems, fast and reliable automated lens finding methods will be essential in the era of large surveys such as LSST, Euclid, and WFIRST. To tackle this challenge, we introduce CMU DeepLens, a new fully automated galaxy-galaxy lens finding method based on Deep Learning. This supervised machine learning approach does not require any tuning after the training step which only requires realistic image simulations of strongly lensed systems. We train and validate our model on a set of 20,000 LSST-like mock observations including a range of lensed systems of various sizes and signal-to-noise ratios (S/N). We find on our simulated data set that for a rejection rate of non-lenses of 99%, a completeness of 90% can be achieved for lenses with Einstein radii larger than 1.4 and S/N larger than 20 on individual g-band LSST exposures. Finally, we emphasize the importance of realistically complex simulations for training such machine learning methods by demonstrating that the performance of models of significantly different complexities cannot be distinguished on simpler simulations. We make our code publicly available at https://github.com/McWilliamsCenter/CMUDeepLens.

Galaxy Zoo: probabilistic morphology through Bayesian CNNs and active learning

“…These transformations are initially random, and are 'learned' by iteratively minimising the difference between predictions and known labels. We refer the reader to LeCun et al (2015) for a brief introduction to CNNs and to Dieleman et al (2015), Lanusse et al (2018), Kim & Brunner (2017) and Hezaveh et al (2017) for astrophysical applications. Early work with CNNs immediately surpassed nonparametric methods in approximating human classifications (Huertas-Company et al 2015;Dieleman et al 2015).…”

Image Simulations for Strong and Weak Gravitational Lensing

“…Many efforts have recently focused on employing techniques from computer vision and machine learning to go beyond traditional approaches such as visual searches of "blue" arcs near "red" galaxies (Diehl et al 2017), goodness of fit examinations after fitting a model to all candidates (Marshall et al 2009;Chan et al 2015), and public science challenges to discover new strong lensing systems in the large datasets. Neural networks have demonstrated to be able to distinguish between simulated lenses and non-lenses (Lanusse et al 2017;Hezaveh et al 2017). Jacobs et al (2019aJacobs et al ( ,b, 2017 have used convolutional neural networks (CNNs, (LeCun et al 1989)) to produce a catalog of galaxy-galaxy strong lenses (including high-redshift systems) using data from the Dark Energy Survey, and Petrillo et al (2017Petrillo et al ( , 2019 have correspondingly found hundreds of candidates in KiDS data.…”

Identifying strong lenses with unsupervised machine learning using convolutional autoencoder

“…The application of CNNs for detecting these GGSL systems has reached a high success rate in binary classification (Jacobs et al 2017;Petrillo et al 2017;Ostrovski et al 2017;Bom et al 2017;Hartley et al 2017;Avestruz et al 2019;Lanusse et al 2018); however, the application of supervised machine learning such as CNNs is prone to human bias and training set bias which may not properly represent the diversity of real GGSL systems observed in future surveys. Additionally, GGSLs are rare events in the Universe so that there is insufficiently homogeneous data for training in supervised machine learning methods.…”