Euclid - Challenge the Machines

Finding Strong Gravitational Lenses

Motivation

Gravitational lenses are useful for putting constraints on values of the Hubble constant, the dark energy equation of state, dark matter halo substructure, and for studying high-redshift galaxies at high spatial and flux magnification. Rare lens systems, for example the Jackpot lens, are particularly useful for dark energy and halo profiles. There is a need to find rare lens configurations to obtain statistics on lens and source parameters and to improve cosmological parameters. However rare lens systems can only be found with large lens samples, and large survey samples are also needed to remove small number statistics. The Euclid telescope, due for launch in 2021, will perform an imaging and slitless spectroscopy survey over half the sky, to map baryon wiggles and weak lensing. During the survey Euclid is expected to resolve 100,000 strong gravitational lens systems. This is ideal to find rare lens configurations, provided they can be identified reliably and on a reasonable timescale.

	An example of a lensed galaxy, blue arcs around a central object.

Problem

The question of how to find strong gravitational lenses in large datasets is a daunting data mining problem within Euclid. Feature recognition by eye, which is currently the best method for identifying strong gravitational lenses, will not be fast enough to cope with the amount of data received. The Strong Lensing group within the Euclid consortium set up the Euclid Strong Lensing challenge. This was a challenge aimed at building machine learning techniques to classify Euclid-like simulations as containing a lens or not. Amazingly, these artificial intelligences outperformed a human expert who trawled through the entire dataset. However we know from other Zooniverse projects, like Spacewarps, that volunteers are excellent at finding gravitational lenses. Can these artificial intelligence machine learning programs out perform all humans?

		Is there a lens in either of the images?

Goal

The goal of this project is to test if humans are still better at identifying gravitational lenses than the machine learning algorithms. Machines have been tested on this same data set, meaning a fair comparison can be made between the two techniques. Once the images have been classified, the data will be analysed to find out if the humans are still better, or have machine learning techniques improved enough to beat the humans. The images to be classified are simulated, designed to simulate images from the Kilo Degree Survey. The size of each image is 10 arc-second by 10 arc-second. The redshift of centre galaxy in each image is 1.5, meaning the actual distance in the image is approximately 84 kiloparsecs. We are using simulated Kilo Degree Survey images as simulations of the expected Euclid images are not yet ready and we expect to get similar results using the Euclid simulations.

App version

As well as the browser version of this project, this project is available to complete using the app version of Zooniverse. The app version allows you to swipe yes or swipe no for each image, speeding up classifying the galaxies (and it is more fun). The Zooniverse app is available for both Android and iOS and can be downloaded for free from the Google Play Store and the App store for Android and iOS respectively.