Astro Seminar: Analysing the SEDs of protoplanetary disks with machine learning

Spectral energy distributions (SEDs) of protoplanetary disks can be used to determine the disk's structural properties and dust composition. However, this analysis is known to be highly degenerate. Therefore, a full Bayesian analysis is required to account for parameter uncertainties and degeneracies. The main challenge here is computational speed, as one proper full radiative transfer model requires at least a couple of CPU-minutes to compute and about 10^6 models are needed for a typical analysis.

To circumvent the computational cost problem, we created neural networks to emulate the SED generation process. Then, we performed a Bayesian analysis on 30 well-known protoplanetary disks to determine the posterior distributions of all parameters. Furthermore, we evaluated the uncertainties from SED analysis and quantify the strongest degeneracies.

During this seminar talk, I’ll summarize the above-mentioned project and conclude with an outlook on my new tasks here in Exeter.