avatar
AstroAI
Developing Artificial Intelligence to Solve the Mysteries of the Universe
  • HOME
  • RESEARCH
  • EARTHAI
  • PEOPLE
  • EVENTS
  • LATEST NEWS
  • LUNCH TALKS
  • SUMMER PROGRAM
  • WORKSHOP
  • APPLY
  • CONTACT
Home AstroAI Workshop 2026
Workshop_abstract2026
Cancel

AstroAI Workshop 2026

Details Invited Speakers Abstracts Register Schedule Venue Accommodations Code of Conduct

Braden Draucek

Towards an Understanding of AGN UV-NIR Spectra Using a Physically Motivated Spectral Decomposition Neural Network

Presenter: Braden Draucek (Western Michigan University)

Title: Towards an Understanding of AGN UV-NIR Spectra Using a Physically Motivated Spectral Decomposition Neural Network

Date/Time: Monday, June 15, 4:00 PM - 5:30 PM

Abstract: Active Galactic Nuclei (AGN) are among the most luminous objects in the universe, and much of this radiated energy emerges within the near IR to far-UV. The bulk of this light is thought to come from a combination of the accretion structure, the broad line region (BLR), and the dusty torus (TOR). The BLR and TOR are thought to be energized by the time-variable flux of ionizing photons emitted by the central accretion structure. Unfortunately, the current state of spectral fitting codes for these regions are fit with phenomenological or mathematical models that incorporate very little of the underlying physics. To improve upon this, I introduce a physically informed neural network to model the composite AGN continuum light spectrum using a simplified, yet physically motivated, model of the accretion structure and TOR, and a completely physical model for the BLR diffuse continuum. Fits to predictions from the photoionization code CLOUDY will be used to help learn constraints on the physical parameters controlling the model for the BLR and TOR. XSPEC will be used to learn constraints for the SED of the accretion structure. The incorporated neural network method is a physically informed neural network that takes shape in a transformer architecture including a multiheaded-attention mechanism that encodes the imported spectrum before running through some dense layers to get the optimal weights of the model through a recursive process. The physical implementation comes from forcing the potential output to fit within known constraints and applying a physically motivated loss function that is built upon known relationships from building up the training set. Current modeling of this continuum is generally not complete by missing some of the regions continua spectra as well as not implementing any form of AI in their methodology. The physically informed neural network will allow for a new method of fitting these spectra as well as be a more complete model for analyzing data from future missions in AGN spectroscopy.

-->

© 2026 AstroAI. Some rights reserved.

Powered by Jekyll with Chirpy theme.

A new version of content is available.