Recently, Long Shen, a researcher at the Lijiang Astronomical Observatory at the Yunnan Observatory of the Chinese Academy of Sciences, teamed up with Professor Er Xinzhong of the Cosmology Research Group of the Southwest Astronomical Research Institute of Yunnan University to discover 38 new candidates for strong gravitational lensing using the method of deep learning of artificial intelligence. The findings are published in the Journal of the Royal Astronomical Society (MNRAS).
Gravitational lensing mechanism diagrams/images are derived from the network.
The galaxy-scale strong gravitational lensing system is an important cosmological probe that can be used to study in depth many scientific issues in cosmology and astrophysics, such as the nature of dark matter, the formation and evolution of galaxies, and the measurement of Hubble constants. However, the number of strong lens systems that have been identified is too small, which restricts the research on related astrophysics.
Images of 4 of the 38 newly discovered strong lens candidates.
How to search for more strong lens samples is the main problem in the current work. Tens of thousands of strong lens systems are expected to be discovered through the next generation of large-scale metering projects. But how to quickly find strong lens candidates in a massive array of celestial images? In recent years, the rapid development of artificial intelligence has provided us with new possibilities. Internationally, the relevant research team has used convolutional neural network method to search for strong gravitational lensing system.
Long-term engaged in the study of deep learning of artificial intelligence, with the erxin team to build and train a convolutional neural network, the neural network using the Julia language according to the characteristics of gravitational lensing data, with small scale, fast speed, targeted characteristics. The researchers used it in the 2.6-meter Sky Patrol (VST) Kilo-Degree Survey-KiDS data at the European Southern Observatory to find 38 new strong lens candidates.
In addition, by testing the performance of convolutional neural networks on different observational conditions and training networks with training sets of different sizes, the researchers tested the stability of convolutional neural networks. The neural network built in the study can also be applied to other survey data.
Long Is the co-author of the paper. The research work is funded by the National Nature Fund project, Yunnan Province overseas high-level talent program, etc.