How do the properties of quasars, such as their black hole mass or growth rate, influence the shape of their spectra? Can we predict the quasars' characteristics from their spectra even with only very limited spectral coverage? In order to understand these questions our group develops new multi-modal generative models for quasar spectra based on Gaussian processes. These models treat every quasar as a vector of latent properties such that the spectrum and all physical properties of the quasar are associated with non-linear functions of those latent parameters. The new machine learning models allow us to study the spectral dependencies of quasar properties, as well as predict the quasars' characteristics from limited spectral data.
How do the properties of quasars, such as their black hole mass or growth rate, influence the shape of their spectra? Can we predict the quasars' characteristics from their spectra even with only very limited spectral coverage? In order to understand these questions our group develops new multi-modal generative models for quasar spectra based on Gaussian processes. These models treat every quasar as a vector of latent properties such that the spectrum and all physical properties of the quasar are associated with non-linear functions of those latent parameters. The new machine learning models allow us to study the spectral dependencies of quasar properties, as well as predict the quasars' characteristics from limited spectral data.