Séminaires : Documents

Parameter estimation in Cosmology with Cobaya and Machine-Learning
Jesus Torrado (Padova University)

I will briefly present Cobaya, an inference framework aimed at model pipelines with complex interdependencies between parameters and intermediate quantities, and use it to illustrate some aspects of classic Bayesian inference in Cosmology. Next, I will discuss the effort by me and my collaborators towards a no-pre-training, drop-in replacement for classical parameter inference on very slow posteriors, using a Gaussian Process surrogate model that actively learns the posterior distribution and converges in O(100) fewer evaluations.