-

AI in the Radio Sky: Data Processing, HI Galaxy Detection, and Beyond

组织者

贾赫德·阿贝迪 , 迪潘詹·德伊 , Puskar Mondal , 亚历杭德罗·托雷斯-奥韦拉

演讲者

梁汝汐

时间

2026年04月08日 14:30 至 15:30

地点

A3-2-301

线上

Zoom 928 682 9093 (BIMSA)

摘要

A fundamental challenge in radio astronomy is the effective extraction of celestial signals, such as neutral hydrogen (HI) galaxies, from data heavily contaminated by radio frequency interference (RFI). This talk begins with an overview of the radio astronomical data processing pipeline. I will then introduce the core principles of relevant machine learning techniques—specifically object detection and image segmentation—explaining how these computer vision frameworks can be adapted for HI galaxy signal identification. A central part of the talk will focus on my previous work: developing a deep learning method based on the Mask R-CNN framework integrated with the PointRend method for HI galaxy detection. This includes a look at the methodology for constructing realistic training datasets, specifically through magnetohydrodynamical simulations (e.g., TNG100) and light-cone stacking to model the HI sky. Finally, the talk will look "beyond" this specific project to discuss broader applications of machine learning in astronomy, including a brief example of its usage in gravitational wave data processing.