Deep learning application for astrophysics: Supermassive black hole, dark matter substructures, and galaxy morphology

Lin, Yao-Yu

Permalink

https://hdl.handle.net/2142/116048 Copy

Description

Title

Deep learning application for astrophysics: Supermassive black hole, dark matter substructures, and galaxy morphology

Author(s)

Lin, Yao-Yu

Issue Date

2022-07-08

Director of Research (if dissertation) or Advisor (if thesis)

Holder, Gilbert

Doctoral Committee Chair(s)

Gammie, Charles F

Committee Member(s)

• Liu, Xin
• Neubauer, Mark

Department of Study

Physics

Discipline

Physics

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Black holes
• Dark Matter: Machine Learning
• Deep Learning
• VLBI, Vision Transformer

Abstract

Modern astronomy research has been thriving due to newly observations. To handle large and novel datasets, deep learning provides new way to tackle the challenges in data analysis. This thesis include deep learning applications on several astrophysics projects related to supermassive black holes (SMBH), dark matter substructures in strong gravitational lensing, and galaxy morphology classification. For SMBH, the Event Horizon Telescope (EHT) recently released the first horizon-scale images of the black hole in M87. Combined with other astronomical data, these images constrain the mass and spin of the black hole as well as the accretion rate and magnetic flux trapped on the black hole. An important question for EHT is how well key parameters such as spin, ring-size and trapped magnetic flux can be extracted from present and future EHT data alone. We explore parameter extraction using a convolutional neural network (CNN) trained on high resolution synthetic images drawn from state-of-the-art simulations. We find that the neural network is able to recover spin and flux with high accuracy. We are particularly interested in interpreting the neural network output and understanding which features are used to identify, e.g., black hole spin. Using feature maps, we find that the network keys on low surface brightness features in particular. We further investigate ring-size estimation using a neural network trained on noisy and blurry images drawn from state-of-the-art simulations. We find that the neural network can recover the physical scale, $GM/(Dc^2)$, in blind data tests of synthetic M87* images with an uncertainty distribution that is approximately Gaussian, with standard deviation around $13\%$. While we found our ML pipeline works well in image domain, we further investigate the possibility of using ML for an end-to-end VLBI data cassification. We propose a data-driven approach to analyze complex visibilities and closure quantities for radio interferometric data with neural networks. Using mock interferometric data, we show that our neural networks are able to infer the accretion state as either high magnetic flux (MAD) or low magnetic flux (SANE), suggesting that it is possible to perform parameter extraction directly in the visibility domain without image reconstruction. We have applied VLBInet to real M87 EHT data taken on four different days in 2017 (April 5, 6, 10, 11), and our neural networks give a score prediction $0.52, 0.4, 0.43, 0.76$ for each day, with an average score $0.53$, which shows no significant indication for the data to lean toward either the MAD or SANE state. This thesis also include three individual projects: Weighing Supermassive Black Holes directly from photometric light curves with Deep Learning, Hunting for Dark Matter Substructures in Strong Gravitational Lensing with Neural Networks, Galaxy Classification with Vision Transformer. We find that deep learning could help tackle these problems, and we also discuss future directions in these projects.

Graduation Semester

2022-08

Type of Resource

Thesis

Copyright and License Information

Copyright 2022 Yao-Yu Lin

Owning Collections