|
||
---|---|---|
figures | ||
poster | ||
preamble | ||
rebuttal | ||
.gitignore | ||
.latexmkrc | ||
README.md | ||
additional.bib | ||
bayesian_warped_dependent_gp.pdf | ||
bayesian_warped_dependent_gp.tex | ||
bayesian_warped_dependent_gp.tex.latexmain | ||
deep_probabilistic_workshop.pdf | ||
deep_probabilistic_workshop.tex | ||
neurips_2018.sty | ||
neurips_2018.tex | ||
references.bib | ||
zotero_export.bib |
README.md
This paper summarizes a way of formulating priors for multiple functions using GPs. The covariance structure between these processes is assumed to be governed by convolutions over latent processes with non-linear alignment of the different outputs. Both the alignment functions and a warping after the convolutions are learned in the context of a deep GP.
Building the Document
The figures in this paper depend on a newish version of the tikz package. Everything starting with TeX Live 2016 should be fine. Additionally, since they use the graphdrawing library, the document must be built using LuaTeX. The easiest way to build the document is by running latexmk
in the main directory - a .latexmkrc
-file is provided with the correct configuration. If you want to use a different way of building the document, please make sure to use lualatex
.