Bo Han Chen(Physics Department, NTHU); Tomotsugu Goto(Institute of Astronomy, NTHU); Seong Jin Kim(Institute of Astronomy, NTHU)
In this project, we develop an efficient active galactic nucleus selection Artificial Intelligence based on Deep Neural-Network model constructing. It is important to do selection between active galactic nucleus (AGNs) and star-forming galaxies(SFGs). However, AGNs are often obscured by gas and dust, and those obscured AGNs tend to be missed in optical, UV and soft X-ray observations. To solve this problem, other parameters of the objects, for example, Mid-IR light can help us to recover them in an obscuration-free way using their thermal emission. By the surely separated AGNs and SFGs samples, we input their photometric data as our training basis for our Deep Neural-Network. Powerful data exploration by deep neural-network offers a promising future in AGNs selection. We trained the whole data basis for 1000 epochs. After that, we applied the trained Neural Network on another test set. This gives us an AGN recovering rate ≅ 70.98% (the proportion of how much of the exact AGNs are discovered) , with a total accuracy ≅ 88.15% (the proportion of how much of the galaxies are correctly justified the galaxy type) . Compared with SED fitting AGN recovering rate ≅ 48% (from Ting-Chi et al. 2017) , we obtained a 23% higher recovering rate. In conclusion, we got a reliable technique on galaxy type classification. This provides us a promising way to estimate AGN fraction on photometric data.