From 29c73bfd055ca2da154be165450021dee89c1c84 Mon Sep 17 00:00:00 2001 From: davidwhogg Date: Sat, 2 Dec 2017 17:01:27 -0500 Subject: [PATCH 1/4] I promised you an Abstract; feel free to ignore --- paper/ms.tex | 30 +++++++++++++++++++++++++++--- 1 file changed, 27 insertions(+), 3 deletions(-) diff --git a/paper/ms.tex b/paper/ms.tex index 60126273..3c91749b 100644 --- a/paper/ms.tex +++ b/paper/ms.tex @@ -45,6 +45,7 @@ \newcommand{\eqalt}[1]{Equation~\eqref{#1}} \newcommand{\eqlabel}[1]{\label{eq:#1}} +\newcommand{\documentname}{\textsl{Article}} \newcommand{\sectionname}{Section} \newcommand{\sectref}[1]{\ref{sect:#1}} \newcommand{\Sect}[1]{\sectionname~\sectref{#1}} @@ -80,9 +81,32 @@ \affiliation{Center for Computational Astrophysics, Flatiron Institute, New York, NY} \begin{abstract}\noindent - - This is an abstract. - +Gaussian processes provide a flexible framework +for modeling nuisances or systematics that are unknown, non-linear functions of known +control parameters. +Examples include stochastic variability (as a function of time) of stars, wavelength-trace +or continuum calibration (as a function of wavelength) in spectrographs, +or dust (as a function of spatial position) in the Milky Way. +In this purely pedagogical \documentname, I explain what Gaussian processes are, +show examples of their use, provide specific advice for implementation, and discuss their +limitations and scope of applicability. +A Gaussian process can be thought of as a prior over a space of functions, +or a function with a formally infinite number of free parameters (a non-parametric model), +or a model for correlated Gaussian noise; +I introduce the processes through the latter point of view, but show how the multiple +views are related by showing that they can be used to deliver posterior beliefs +over continuous functions that explain noisy data. +I connect the idea of the Gaussian process to ideas (common in +data analysis) of simultaneously fitting and marginalizing out additive nuisances. +Most uses of Gaussian process in astronomy have been in the context of functions of one +or a few variables (time, or space, or wavelength); I emphasize that there might be +valuable applications in which the ambient dimension is far larger (for example, to model +nonlinear functions of thousands of stellar lightcurves), and show some examples. +In certain limits, Gaussian processes can be physically motivated as the response +of a linear system to Gaussian white-noise forcing, which makes them interesting for +modeling finite-Q oscillations in, for example, asteroseismology; I give some examples. +Finally, I discuss ways to make Gaussian processes fast, by exploiting problem structure, +and point to relevant literature and software. \end{abstract} \keywords{% From 7b306bc8f1a28ba5f9b365f504c9e9c27c3a2f25 Mon Sep 17 00:00:00 2001 From: davidwhogg Date: Sat, 2 Dec 2017 17:03:29 -0500 Subject: [PATCH 2/4] shorts --- paper/ms.tex | 3 +++ 1 file changed, 3 insertions(+) diff --git a/paper/ms.tex b/paper/ms.tex index 3c91749b..da59c4e8 100644 --- a/paper/ms.tex +++ b/paper/ms.tex @@ -67,6 +67,9 @@ \newcommand{\response}[1]{{\color{blue}#1}} +\shorttitle{gaussian processes in astronomy} +\shortauthors{foreman-mackey} + \begin{document} \sloppy\sloppypar\raggedbottom\frenchspacing From 6ad3278243cb2bb4170db13a3f562d3ba862078e Mon Sep 17 00:00:00 2001 From: davidwhogg Date: Sat, 2 Dec 2017 17:06:02 -0500 Subject: [PATCH 3/4] tweak --- paper/ms.tex | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/paper/ms.tex b/paper/ms.tex index da59c4e8..c137f87a 100644 --- a/paper/ms.tex +++ b/paper/ms.tex @@ -104,7 +104,7 @@ Most uses of Gaussian process in astronomy have been in the context of functions of one or a few variables (time, or space, or wavelength); I emphasize that there might be valuable applications in which the ambient dimension is far larger (for example, to model -nonlinear functions of thousands of stellar lightcurves), and show some examples. +nonlinear functions of thousands of elements of housekeeping data), and show some examples. In certain limits, Gaussian processes can be physically motivated as the response of a linear system to Gaussian white-noise forcing, which makes them interesting for modeling finite-Q oscillations in, for example, asteroseismology; I give some examples. From e927d0c50a4e0b3a2e6aed2e66b0c4cd011207a1 Mon Sep 17 00:00:00 2001 From: davidwhogg Date: Sat, 2 Dec 2017 17:09:13 -0500 Subject: [PATCH 4/4] keywords --- paper/ms.tex | 16 +++++++--------- 1 file changed, 7 insertions(+), 9 deletions(-) diff --git a/paper/ms.tex b/paper/ms.tex index c137f87a..9aee5cca 100644 --- a/paper/ms.tex +++ b/paper/ms.tex @@ -113,15 +113,13 @@ \end{abstract} \keywords{% - %methods: data analysis - %--- - %methods: statistical - %--- - %asteroseismology - %--- - %stars: rotation - %--- - %planetary systems + asteroseismology + --- + methods: data analysis + --- + methods: statistical + --- + methods: numerical } \section{Introduction}