Skip to content

Commit

Permalink
Add czs 2023 slides
Browse files Browse the repository at this point in the history
  • Loading branch information
@kazewong
kazewong committed Aug 13, 2023
1 parent bd1e6b3 commit 7446d74
Show file tree
Hide file tree
Showing 8 changed files with 427 additions and 0 deletions.
1 change: 1 addition & 0 deletions slides/amaldi2023/intro.html
View file
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,7 @@ <h4 class="header">AI hype</h4>
</div>
</section>


<section>
<h2 class="header">What can AI do for GW?</h4>
<p class="fragment" style="color: rgb(100,255,100); font-size: xxxx-large; text-align: left;"> Illustrated with inference problems</p>
Expand Down
Empty file removed slides/csz2023/top.html
View file
Empty file.
91 changes: 91 additions & 0 deletions slides/czs2023/SBI.html
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,91 @@
<section>
<h2 class="header">Simulation-based inference</h2>
</section>

<section>
<h4 class="header">Simulate, train, compare</h4>
<div class="column">
<img class="fragment" data-src="https://dark.nbi.ku.dk/news/dark-news-21/simulation-based-inference-of-dynamical-galaxy-cluster-masses-with-3d-convolutional-neural-networks/Doogesh.png" width="100%">
<img class="fragment" data-src="https://ep-news.web.cern.ch/sites/default/files/inline-images/EP-March-2021/ML-EP-news3.png" width="100%">

</div>
</section>

<!-- Single event inference -->

<section>
<h4 class="header"> Single GW event inference </h4>
<div class="row">
<div class="fragment" style="padding-right: 20px">
<p style="padding:0%; margin:0%;"> Data </p>
<img data-src="../scma8/images/TimeSeries.png" height="360px">
</div>
<div class="fragment" style="padding-right: 20px">
<p style="padding:0%; margin:0%;"> Source properties </p>
<img data-src="../scma8/images/m1m2PE.png" height="360px">
</div>
</div>
</section>

<section>
<h4 class="header ">Neural posterior estimation for gravitational wave</h4>
<div class="row">
<blockquote class="twitter-tweet fragment"><p lang="en" dir="ltr">Can you trust deep learning for scientific inference? And what can you do when results are inaccurate? We address these questions for ML-based inference of complex gravitational wave models and get highly accurate and reliable results. <a href="https://t.co/MLC4HDAv1q">https://t.co/MLC4HDAv1q</a> 1/12 <a href="https://t.co/sK6vJhvWjh">pic.twitter.com/sK6vJhvWjh</a></p>&mdash; Max Dax (@maximilian_dax) <a href="https://twitter.com/maximilian_dax/status/1588177410209742848?ref_src=twsrc%5Etfw">November 3, 2022</a></blockquote> <script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>
<div class="column fragment" style="margin-left: 100px;">
<img data-src="https://d3i71xaburhd42.cloudfront.net/284498ac6472eb973ef9bbdd97d2bfb667b7f652/2-Figure1-1.png" width="60%" style="aspect-ratio: auto;">
<img data-src="https://d3i71xaburhd42.cloudfront.net/284498ac6472eb973ef9bbdd97d2bfb667b7f652/14-Figure10-1.png" width="60%" style="aspect-ratio: auto;">

</div>

</div>

</section>

<!-- Unmodelled search-->

<section>
<h4 class="header">Population inference</h4>
<div class="row">
<div class="column fragment">
<img data-src="https://d3i71xaburhd42.cloudfront.net/85920220d916aa96e5ef7f4f8eeca6152499a235/3-Figure1-1.png" width="100%">
</div>
<div class="column fragment">
<img data-src="https://d3i71xaburhd42.cloudfront.net/85920220d916aa96e5ef7f4f8eeca6152499a235/4-Figure3-1.png" width="100%">
</div>
</div>
<p style="color: white;" class="fragment">If you believe us, then most of the bbhs are from cluster!<span class="fragment" style="font-size: medium; color: tomato; text-align: center; ">(I don't)</span></p>
</section>

<section>
<h3>Let's take a moment and think</h3>
<h4 class="fragment"> why the first example is more believable?</h4>
</section>

<section>
<h3 class="header">Lesson from two SBI scenarios</h4>
<div class="column">
<div class="row" style="justify-content: space-between;">
<div class="column" style="color: skyblue;">
<h4 class="fragment" style="color: skyblue;">Perks of SBI</h4>
<ol>
<li class="fragment">Fast inference</li>
<li class="fragment">Low retuning human effort</li>
<li class="fragment">Flexible assumptions</li>
</ol>
</div>
<div class="column" style="color: tomato;">
<h4 class="fragment" style="color: tomato;">Dangers of SBI</h4>
<ol>
<li class="fragment">"Why is my mass negative"</li>
<li class="fragment">Are we really winning?</li>
<li class="fragment">Retraining cost</li>
</ol>
</div>
</div>
<div class="column" style="margin-top: 30px;">
<p class="fragment" style="margin: 3px; color: skyblue;">Fixed model, changing data -> Good </p>
<p class="fragment" style="margin: 3px; color: tomato;">Fixed data, change model -> Bad </p>
</div>
</div>

</section>
129 changes: 129 additions & 0 deletions slides/czs2023/hybrid.html
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,129 @@
<section>
<h2 class="header">Hybrid method</h2>
</section>

<!-- Differentiable models -->

<section>
<h4 class="header">Infrastructure around AI is just as cool</h4>
<div class="column">
<div class="row" style="justify-content: space-evenly;">
<img class="row fragment" data-src="../scma8/images/Jax.png">
<ol class="bright fragment">
<li>Autodiff</li>
<li>JIT compilation</li>
<li>Simple vectorization</li>
<li>GPU with XLA</li>
</ol>
</div>
<div class="row fragment" style="justify-content: space-evenly;">
<img class="row" data-src="../scma8/images/jax_grad.png" width="360px;" style="padding-right: 20px;">
<img class="row" data-src="../scma8/images/jax_compile.png" width="360px;">
</div>
</div>
</section>

<section>
<div class="row">
<h4 class="header"> Ripple - GW model in Jax</h4>
<img data-src="../scma8/images/ripple_github_tag.png" style="margin-top: 0; margin-left: 20px;">
</div>
<div class="column">
<img data-src="../scma8/images/Ripple_benchmark.png" height="360px;">
<div class="row fragment">

<span style="color: rgb(100,255,100); margin: 0;">\[\sim 10^6 \]</span>
<span style="color: rgb(100,255,100);">likelihood in a second, that's a whole run!</span>
</div>
</div>
</section>

<!-- flowMC -->

<section>
<div class="row">
<h4>flowMC - Normalizing flow enhanced MCMC</h4>
</div>
<div class="row">
<div class="Column">
<img data-src="../scma8/images/flowMC.png" width="400px">
<ol>
<li class="fragment">Black box interface</li>
<li class="fragment">GPU+XLA support</li>
<li class="fragment">Gradient-base sampler</li>
<li class="fragment">Powerful flow model</li>
</ol>
<p class="fragment" style="color: rgb(100,255,100); font-size:medium"> Worked for exoplanet, protein structures, and more!</p>
<img class="fragment" src="../scma8/images/flowMC_github_tag.png" width="260px">

</div>

<div class="column fragment">
<iframe src="https://flowmc.readthedocs.io/en/main/" style="background: #FFFFFF; margin-top: 20px;" title="flowMC-doc" width=1080 height=560></iframe>
</div>
</div>
</section>

<section>
<h4 class="header">How does it work? Normal MCMC</h4>
<video controls autoplay>
<source src="../amaldi2023/videos/NormalMCMC.mp4" type="video/mp4">
</video>
</section>

<section>
<h4 class="header">How does it work? Building normalizng flow</h4>
<video controls autoplay>
<source src="../scma8/videos/BuildingFlow.mp4" type="video/mp4">
</video>
</section>

<section>
<h4 class="header">How does it work? Proposal with normalizing flow</h4>
<video controls autoplay>
<source src="../scma8/videos/FlowMC.mp4" type="video/mp4">
</video>
</section>

<!-- Jim -->

<section>
<div class="row">
<h4 class="header"> Putting the tools together - Jim</h4>
<img data-src="../scma8/images/jim_github_tag.png" style="margin-top: 0; margin-left: 20px;">
</div>
<div class="column">
<div class="row">
<div class="fragment">
<p> Real data </p>
<img class="row" data-src="../scma8/images/jim_posterior.png" height="400px;">
</div>
<div class="fragment">
<p> Simulated population</p>
<img class="row" data-src="../scma8/images/jim_pp_plot.png" height="400px;">
</div>
</div>
<p style="color: rgb(100,255,100); font-size: x-large; margin-left:-80px" class="fragment"> 2 minutes on GPU per event</p>
<p style="color: rgb(100,255,100); font-size: x-large; margin-left:-80px; margin-top: 3px;" class="fragment"> I am almost done cleaning it up, expect news later this month or in August!</p>
</div>
</section>

<section>
<h4 class="header">Preliminary development - going beyond Gaussian</h4>
<div class="row" style="justify-content: space-evenly;">
<div class="column fragment">
<p>Lensing with cosmic ray</p>
<img data-src="https://s3.amazonaws.com/aasie/images/2041-8205/949/2/null/apjlacd645f2_hr.jpg" height="400px;">
</div>
<div class="column fragment">
<p>GW with glitch (not yet)</p>
<img data-src="../amaldi2023/images/slic_corner.png" height="400px;">
</div>
</div>
</section>

<style>
.katex-display{
margin: 0px;
}
</style>
78 changes: 78 additions & 0 deletions slides/czs2023/intro.html
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,78 @@
<section>
<h2> Success and failure of ML in gravitational wave </h2>
<h4> Kaze Wong </h4>
<div style="position: fixed; bottom: 0; right: -600;">
<img data-src="../../asset/images/CCA_logo_color.png" width="20%"/>
</div>
<!-- <img style="position: fixed; bottom: 0; right: 800;" data-src="./images/qrcode_amaldi1.png" width="30%"/> -->
</section>

<section>
<div class="column">
<h4 class="header"> O4 has started recently</h4>
<blockquote class="twitter-tweet tw-align-center" height="360">
<p lang="en" dir="ltr">Welcome to the fourth observing run! <a href="https://twitter.com/hashtag/O4IsHere?src=hash&amp;ref_src=twsrc%5Etfw">#O4IsHere</a><br><br>O4 will last 20 months, with <a href="https://twitter.com/ego_virgo?ref_src=twsrc%5Etfw">@ego_virgo</a> and <a href="https://twitter.com/KAGRA_PR?ref_src=twsrc%5Etfw">@KAGRA_PR</a> joining us in the run. We expect the detection rate will be roughly twice in our past observing run<br><br>More news linked from <a href="https://t.co/kHnyuP8I2I">https://t.co/kHnyuP8I2I</a> <a href="https://twitter.com/hashtag/ObservingRun4?src=hash&amp;ref_src=twsrc%5Etfw">#ObservingRun4</a> <a href="https://t.co/B9fhEG96r0">pic.twitter.com/B9fhEG96r0</a></p>&mdash; LIGO (@LIGO) <a href="https://twitter.com/LIGO/status/1661388186042417152?ref_src=twsrc%5Etfw">May 24, 2023</a>
</blockquote>
<script async src="https://platform.twitter.com/widgets.js" charset="utf-8">
</script>
</div>
</section>

<section>
<div class="column">
<h4 class="header"> Events are coming in hot!</h4>
<iframe src="https://gracedb.ligo.org/superevents/public/O4/" style="background: #FFFFFF;" title="Gracedb" width=1080 height=560></iframe>
</div>
</section>

<section>
<h4 class="header">The future is on the horizon</h4>
<div class="row" style="justify-content: space-evenly;">
<div class="r-stack">
<div class="fragment" data-fragment-index="2">
<img data-src="../scma8/images/3GHorizon.png" width="360px">
</div>
<div class="column fragment fade-in-then-out" data-fragment-index="1">
<img data-src="../scma8/images/ET.jpg" width="360px">
<img data-src="../scma8/images/CE.png" width="360px">
</div>
</div>
<div class="r-stack">
<div class="fragment" data-fragment-index="2">
<img data-src="../scma8/images/LISA_detector.png" width="480px">
</div>
<div class="fragment fade-in-then-out" data-fragment-index="1">
<img data-src="../scma8/images/LISA.png" width="480px">
</div>
</div>
</div>
</section>

<section>
<h4 class="header"> Machine learning in GW</h4>
<div class="r-stack">
<div class="fragment fade-in-then-out">
<div class="column">
<p style="padding:0%; margin:0%;"> Search </p>
<img data-src="../scma8/images/GravitySpy.png" width="720px">
</div>
</div>
<div class="fragment fade-in-then-out">
<div class="column">
<p style="padding:0%; margin:0%;"> Simulation </p>
<iframe width="640px" height="480px" src="https://www.youtube.com/embed/c-2XIuNFgD0" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen></iframe>
</div>
</div>
<div class="fragment fade-in">
<p style="padding:0%; margin:0%;"> Inference </p>
<img data-src="../scma8/images/InferenceMeme.jpg" width="720px">
</div>
</div>
</section>

<section>
<h4 class="header">This talk is more about opinions than methods</h4>
<p class="fragment header" style="color: rgb(100,255,100)">"Should I use this method for this problem?"</p>
<p class="fragment header" style="color: rgb(255,100,100)">"Should I really approach the problem like this?"</p>
</section>

50 changes: 50 additions & 0 deletions slides/czs2023/model.html
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,50 @@
<section>
<h2 class="header">Model design</h2>
</section>

<section>
<h4 class="header"> Challenges for population analysis </h4>
<div class="row">
<img data-src="https://s3.amazonaws.com/aasie/images/2041-8205/913/1/L7/apjlabe949f3_hr.jpg" height="480px;">
<div class="column">
<ol style="margin:50px; padding-left:20px; margin-bottom: 0px;">
<li class="fragment"> Growingly difficult to build models </li>
<li class="fragment"> Hard to go to high dimension</li>
<li class="fragment"> Powered by smart people (short in supply) </li>
</ol>
<img class="fragment" data-src="https://d3i71xaburhd42.cloudfront.net/52d01218711c55ad9c8da71dee9a663bcc0b9330/22-Figure11-1.png" height="200px;" style="margin-left:20px; margin-top:42px;">
</div>
</div>
</section>

<!-- Flexible model -->

<section>
<h4 class="header">Flexible population model</h4>

<div class="column fragment fade-in">
<img data-src="../scma8/images/David_money_plot.png">
<p class="fragment" style="color: rgb(224, 71, 71); font-size:24pt"> Okay it describes the data now, but I have no clue what does it mean.</p>
</div>
</section>

<!-- Symbolic regression -->

<section>
<div class="row">
<h4 style="margin-left:27%"> Symbolic regression </h4>
<img data-src="../scma8/images/PySR.png" style="margin-top: 0; margin-left: 20px;" height="50px">
</div>

<iframe src="https://user-images.githubusercontent.com/7593028/238191929-c8511a49-b408-488f-8f18-b1749078268f.mp4" width="100%" height="480px" frameborder="0" allowfullscreen></iframe>

</section>


<section>
<h4 class="header"> Closing the loop</h4>
<div class="column">
<img class="row" data-src="../scma8/images/Miles_ICML.png">
</div>
</section>

16 changes: 16 additions & 0 deletions slides/czs2023/outro.html
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,16 @@
<section>
<h4 class="header">Take away</h4>
<ul>
<li class="fragment">Ask what do you lose as well</li>
<li class="fragment">Don't reinvent the entire wheel</li>
<li class="fragment">Think about automation and scalability</li>
</ul>
</section>

<section>
<h4 class="header">Discussion time</h4>
</section>

<section>
<div style='position: relative; padding-bottom: 56.25%; padding-top: 35px; height: 0; overflow: hidden;'><iframe sandbox='allow-scripts allow-same-origin allow-presentation' allowfullscreen='true' allowtransparency='true' frameborder='0' height='315' src='https://www.mentimeter.com/app/presentation/aliysb43efr2ok2g4dt7g1f4jp81g4wy/embed' style='position: absolute; top: 0; left: 0; width: 100%; height: 100%;' width='420'></iframe></div>
</section>
Loading

0 comments on commit 7446d74

Please sign in to comment.