Since their earliest days, humans have gazed with wonder upon the firmaments and sought to understand the secrets of the heavenly canopy. In the late 20th century, the cosmological phenomenon known as gravitational lensing developed into one of the primary tools to learn about the distribution of masses in the universe. Predicted by of Albert Einstein’s general theory of relativity, the effect occurs when a large mass bends the path of background light as it travels toward the observer.
Scientists study these distortions in space time to discover exoplanets and resolve deep mysteries about the fundamental nature of the universe, but analyzing just one lens event — a process that involves comparing complex computer simulations with real-world observations — can take weeks or longer.
In what what could prove a major advance for astrophysics, scientists from the Department of Energy’s SLAC National Accelerator Laboratory and Stanford University have developed a neural network model that can analyze images in under a second, some 10 million times faster than traditional methods. They validated the new technique using real images of strongly lensed galaxies taken with NASA’s Hubble Space Telescope.
“Analyses that typically take weeks to months to complete, that require the input of experts and that are computationally demanding, can be done by neural nets within a fraction of a second, in a fully automated way and, in principle, on a cell phone’s computer chip,” said postdoctoral fellow Laurence Perreault Levasseur, a co-author of a study published in Nature (and reported on by SLAC).
The breakthrough comes at time when thousands of new lenses are expected to be discovered as a result of emerging ground and space surveys.
Researchers trained four deep convolutional neural networks — three publicly available models and one they built themselves — with half a million simulated images of gravitational lenses.
After only a day of training, the networks were able to determine the properties of news lenses, such as how its mass was distributed and the degree of image magnification, with an accuracy comparable to previous methods.
In the video below, Phil Marshall of the Kavli Institute of Particle Astrophysics and Cosmology at SLAC/Stanford University explains the optical principles of gravitational lensing using a wineglass.