Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
My research talk for CCA promotion
Search
Dan Foreman-Mackey
February 03, 2022
Science
1
760
My research talk for CCA promotion
A summary of what I've been up to for the past few years and where my research program is going.
Dan Foreman-Mackey
February 03, 2022
Tweet
Share
More Decks by Dan Foreman-Mackey
See All by Dan Foreman-Mackey
Open software for Astronomical Data Analysis
dfm
0
120
Open Software for Astrophysics, AAS241
dfm
2
510
Astronomical software
dfm
1
710
emcee-odi
dfm
1
640
Exoplanet population inference: a tutorial
dfm
3
440
Data-driven discovery in the astronomical time domain
dfm
6
700
TensorFlow for astronomers
dfm
6
770
How to find a transiting exoplanets
dfm
1
460
Long-period transiting exoplanets
dfm
1
310
Other Decks in Science
See All in Science
ACL読み会2024@名大 REANO: Optimising Retrieval-Augmented Reader Models through Knowledge Graph Generation
takuma_matsubara
0
170
Valuable Lessons Learned on Kaggle’s ARC AGI LLM Challenge (PyDataGlobal 2024)
ianozsvald
0
260
白金鉱業Meetup Vol.16_【初学者向け発表】 数理最適化のはじめの一歩 〜身近な問題で学ぶ最適化の面白さ〜
brainpadpr
10
2k
論文紹介: PEFA: Parameter-Free Adapters for Large-scale Embedding-based Retrieval Models (WSDM 2024)
ynakano
0
230
Machine Learning for Materials (Challenge)
aronwalsh
0
250
SciPyDataJapan 2025
schwalbe10
0
150
【健康&筋肉と生産性向上の関連性】 【Google Cloudを企業で運用する際の知識】 をお届け
yasumuusan
0
500
2024-06-16-pydata_london
sofievl
0
620
サメのはなし / How Sharks are born
naospon
0
2.5k
ほたるのひかり/RayTracingCamp10
kugimasa
1
570
事業会社における 機械学習・推薦システム技術の活用事例と必要な能力 / ml-recsys-in-layerx-wantedly-2024
yuya4
4
310
03_草原和博_広島大学大学院人間社会科学研究科教授_デジタル_シティズンシップシティで_新たな_学び__をつくる.pdf
sip3ristex
0
230
Featured
See All Featured
[RailsConf 2023 Opening Keynote] The Magic of Rails
eileencodes
28
9.3k
Java REST API Framework Comparison - PWX 2021
mraible
29
8.5k
Gamification - CAS2011
davidbonilla
80
5.2k
Build The Right Thing And Hit Your Dates
maggiecrowley
34
2.6k
The Myth of the Modular Monolith - Day 2 Keynote - Rails World 2024
eileencodes
22
2.6k
Faster Mobile Websites
deanohume
306
31k
Statistics for Hackers
jakevdp
797
220k
Optimising Largest Contentful Paint
csswizardry
34
3.1k
Designing Experiences People Love
moore
140
23k
How to Think Like a Performance Engineer
csswizardry
22
1.4k
CoffeeScript is Beautiful & I Never Want to Write Plain JavaScript Again
sstephenson
160
15k
Docker and Python
trallard
44
3.3k
Transcript
BUILDING THE SOFTWARE INFRASTRUCTURE FOR ASTROPHYSICS by Dan Foreman-Mackey
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 astrophysics 2
why?
credit: Adrian Price-Whelan / / data: SAO/NASA ADS
my open source contributions 3
None
gaussian processes 4
p(data|physics)
data ~ N(model; noise)
°0.6 °0.3 0.0 0.3 0.6 raw [ppt] 0 5 10
15 20 25 time [days] °0.30 °0.15 0.00 de-trended [ppt] N = 1000 reference: DFM+ (2017)
°0.6 °0.3 0.0 0.3 0.6 raw [ppt] 0 5 10
15 20 25 time [days] °0.30 °0.15 0.00 de-trended [ppt] N = 1000 reference: DFM+ (2017)
data ~ N(model; noise)
data ~ N(model; noise)
so. why not?
data ~ N(model; noise)
None
reference: Ambikasaran, DFM+ (2015)
None
reference: Ambikasaran, DFM+ (2015)
reference: DFM, Agol, Ambikasaran, Angus (2017); DFM (2018); DFM, Luger,
et al. (2021)
None
reference: Gordon, Agol, DFM (2020)
what’s next?
None
None
None
credit: Quang Tran
reference: Luger, DFM, Hedges (2021)
probabilistic inference 5
p(data|physics)
have: physics = > data
want: data = > physics
integral of the form f(physics) p(physics|data) dphysics
None
number of parameters patience required a few tenish not outrageously
many reference: DFM (priv. comm.)
number of parameters patience required emcee a few tenish not
outrageously many reference: DFM (priv. comm.)
number of parameters patience required emcee a few tenish not
outrageously many how things should be reference: DFM (priv. comm.)
None
None
None
None
gradients!
dp(data|physics) / dphysics
automatic differentiation aka “backpropagation”
your model is just code
apply the chain rule
apply the chain rule over and over again . .
.
sounds silly?
it's not! (mostly)
None
None
what’s next?
None
jax.readthedocs.io
my approach to open source 6
None
[1] don’t underestimate users [2] build libraries, not (just) scripts
[3] teach by example
None
None
None
bringing open source practices to research more generally
None
None
None
None
what’s next? 7
7 [1] inference with stochastic or intractable models [2] what
can we do to better support open source in astrophysics
7
7 credit: Adrian Price-Whelan
many fundamental software packages have a shockingly small number of
maintainers.
a selection of some* CCA-supported software: * my apologies for
neglecting your favorites!
None
BUILDING THE SOFTWARE INFRASTRUCTURE FOR ASTROPHYSICS @ CCA