A summary of what I've been up to for the past few years and where my research program is going.
BUILDINGTHESOFTWAREINFRASTRUCTUREFORASTROPHYSICSby Dan Foreman-Mackey
View Slide
who am I?/ /what’ve I been up to?1
7[1] solving Hard™ data analysis problems[2] enabling and empowering astrophysicists
implementation.
data= >physics
open source software for astrophysics2
why?
credit: Adrian Price-Whelan/ /data: SAO/NASA ADS
my open source contributions3
gaussian processes4
p(data|physics)
data ~ N(model; noise)
°0.6°0.30.00.30.6raw [ppt]0 5 10 15 20 25time [days]°0.30°0.150.00de-trended [ppt]N = 1000reference: DFM+ (2017)
so. why not?
reference: Ambikasaran, DFM+ (2015)
reference: DFM, Agol, Ambikasaran, Angus (2017); DFM (2018); DFM, Luger, et al. (2021)
reference: Gordon, Agol, DFM (2020)
what’s next?
credit: Quang Tran
reference: Luger, DFM, Hedges (2021)
probabilistic inference5
have:physics= >data
want:data= >physics
integral of the formf(physics) p(physics|data) dphysics
number of parameterspatience requireda few tenish not outrageously manyreference: DFM (priv. comm.)
number of parameterspatience requiredemceea few tenish not outrageously manyreference: DFM (priv. comm.)
number of parameterspatience requiredemceea few tenish not outrageously manyhow things should bereference: DFM (priv. comm.)
gradients!
dp(data|physics) / dphysics
automatic differentiationaka “backpropagation”
your model is just code
apply the chain rule
apply the chain ruleover and over again. . .
sounds silly?
it's not! (mostly)
jax.readthedocs.io
my approach to open source6
[1] don’t underestimate users[2] build libraries, not (just) scripts[3] teach by example
bringing open source practicesto research more generally
what’s next?7
7[1] inference with stochastic orintractable models[2] what can we do to better supportopen source in astrophysics
7
7credit: Adrian Price-Whelan
many fundamental software packageshave a shockingly small number ofmaintainers.
a selection of some* CCA-supported software:* my apologies for neglecting your favorites!
BUILDINGTHESOFTWAREINFRASTRUCTUREFORASTROPHYSICS @ CCA