May 15 – 17, 2026
College of Hakka Studies at NYCU, Zhubei, Hsinchu County 國立陽明交通大學客家學院(竹北六家校區)
Asia/Taipei timezone

Unsupervised Classification of M87* Polarization Images via Autoencoders

May 16, 2026, 11:45 AM
15m
International Conference Hall, College of Hakka Studies, NYCU 國立陽明交通大學客家文化學院國際會議廳

International Conference Hall, College of Hakka Studies, NYCU 國立陽明交通大學客家文化學院國際會議廳

Speaker

Chieh-Yu Kuo (NTNU)

Description

Polarized emission from the vicinity of black hole systems carries essential information about local magnetic field configurations in the strong gravity regime. In this work, we employ unsupervised learning to a library of model GRMHD Stokes images of M87*, cluster the images based on the polarized image features, and explore how the clustering depends on the model parameters such as black hole spin, accretion type, and electron energies. To perform the clustering, we apply an autoencoder for dimension reduction of the image library, and group the resulting distributions in the latent space with k-mean clustering. The clustering results depend on whether the polarized properties are included in the channel of the input data, implying how the information of model parameters are embedded in different polarized properties.

Participate the oral/poster presentation award competition Yes

Author

Co-author

Prof. Hung-Yi Pu (NTNU,ASIAA)

Presentation materials

There are no materials yet.