The strong gravitational lens finding challenge

Abstract

Large-scale imaging surveys will increase the number of galaxy-scale strong lensing candidates by maybe three orders of magnitudes beyond the number known today. Finding these rare objects will require picking them out of at least tens of millions of images, and deriving scientific results from them will require quantifying the efficiency and bias of any search method. To achieve these objectives automated methods must be developed. Because gravitational lenses are rare objects, reducing false positives will be particularly important. We present a description and results of an open gravitational lens finding challenge. Participants were asked to classify 100,000 candidate objects as to whether they were gravitational lenses or not with the goal of developing better automated methods for finding lenses in large data sets. A variety of methods were used including visual inspection, arc and ring finders, support vector machines (SVM) and convolutional neural networks (CNN). We find that many of the methods will be easily fast enough to analyse the anticipated data flow. In test data, several methods are able to identify upwards of half the lenses after applying some thresholds on the lens characteristics such as lensed image brightness, size or contrast with the lens galaxy without making a single false-positive identification. This is significantly better than direct inspection by humans was able to do. Having multi-band, ground based data is found to be better for this purpose than single-band space based data with lower noise and higher resolution, suggesting that multi colour data is crucial. Multi-band space based data will be superior to ground based data. The most difficult challenge for a lens finder is differentiating between rare, irregular and ring-like face-on galaxies and true gravitational lenses. The degree to which the efficiency and biases of lens finders can be quantified largely depends on the realism of the simulated data on which the finders are trained.Article number, page 1 of 26

Identifying strong lenses with unsupervised machine learning using convolutional autoencoder

“…To further compare the performance of our technique with other supervised machine learning methods and human inspection, we revisit the Strong Gravitational Lens Finding Challenge (Lens Finding Challenge) (Metcalf et al 2019). The final challenge testing data in the Lens Finding Challenge includes 100,000 images which are ∼60 percent of nonlensing images and ∼40 percent of lensing images.…”

“…The strong lensing data are from the Strong Gravitational Lens Finding Challenge (Lens Finding Challenge) (Metcalf et al 2019). The generation of mock images follows the procedures described in Grazian et al (2004) and Meneghetti et al (2008), and starts with a cosmological N-boby simulation, the Millennium simulation (Boylan-Kolchin et al 2009).…”

“…The background objects are modeled by the sources from the Hubble Ultra Deep Field (UDF). The detail of the simulation setup can be found in Metcalf et al (2019).…”

Image Simulations for Strong and Weak Gravitational Lensing

“…Jacobs et al (2019b) use the LENSPOP (Collett 2015) code to generate a training set that consists of hundreds of thousands of labeled simulated examples to train a CNN that classifies lenses and non-lenses. Metcalf et al (2019) use N-body (Millenium) and ray-tracing (GLAMER, ; ) simulations to analyze a variety of methods including CNN's, visual inspection, and arc finders to assess their efficiency and completeness, and identify biases in the face of large future datasets.…”

Surveying the reach and maturity of machine learning and artificial intelligence in astronomy

“…Deep neural nets trained on simulations have resulted in lens discoveries in survey data including the KiDS (de Jong et al, ) by Petrillo et al (, ); and the Dark Energy Survey (Dark Energy Survey Collaboration et al, ) by Jacobs, Collett, Glazebrook, Buckley‐Geer, et al () and Jacobs, Collett, Glazebrook, McCarthy, et al (). A strong lens finding challenge was recently conducted using simulated data (Metcalf et al, ), and deep learning‐based methods outperformed all other methodologies including examination by human experts. Gravitational wave astronomy The recent detection of gravitational wave signals from coalescing black hole binaries (Abbott et al, ), and other related compact systems, has relied on real‐time computation and analysis of streams of data from the Advanced Laser Interferometer Gravitational‐Wave Observatory (LIGO) detectors (Harry and LIGO Scientific Collaboration, ). By incorporating machine learning, Powell et al () improved performance in distinguishing between sources and noise signals, along with reducing the latency of the detection pipeline.…”