ready to resubmit; new latent-variable model

Combined with effort over the weekend, today I finished my revisions (in response to referee) for the MCMC paper I have written with Dan Foreman-Mackey (Flatiron). The paper is on arXiv but we will update it to the new version if it gets accepted. It is a relief to get it done. And the referee comments were very constructive and valuable. What amazes me is that the AAS Journals are willing to publish such an odd paper. I appreciate it, though: The AAS Journals are great journals.

I got a moment in with Christina Eilers (MPIA) to propose (yet another) latent-variable model for stellar spectra. The idea is that stellar spectra are very simple, so the relationship between the latent variables and the spectra could be linear! The relationship between the spectra and the parameters of interest is definitely non-linear, so we use a Gaussian Process to model the relationship between the latents and the labels. It is a hybrid linear–GP latent-variable model, or HLGPLVM! Oh now that's a great name.

No comments:

Post a Comment