Speaker
Le Zhang
(中山大学)
Description
In radio astronomy, reconstructing a sky map from time ordered data (TOD) is an inverse problem. Map-making techniques and gridding algorithms are commonly used, but they have limitations such as computational inefficiency, numerical instability, and an inability to remove beam effects. To address these issues, this study proposes a novel solution using the conditional invertible neural network (cINN) for efficient sky map reconstruction. By training the neural network with forward modeling, it can accurately reconstruct sky maps from simulated TODs. Using FAST as an example, cINN achieves remarkable performance. The reconstruction errors for each pixel can also be accurately quantified by sampling in the latent space of cINN.
Primary author
Le Zhang
(中山大学)
