Skip to content

Commit

Permalink
added zerograds colab
Browse files Browse the repository at this point in the history
  • Loading branch information
@mfischer-ucl
mfischer-ucl committed May 13, 2024
1 parent 9005fca commit e5944a6
Show file tree
Hide file tree
Showing 4 changed files with 12 additions and 36 deletions.
26 changes: 1 addition & 25 deletions .idea/workspace.xml
View file

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

10 changes: 5 additions & 5 deletions _pages/zerograds.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -70,9 +70,9 @@ ol > li::marker {
<h4><strong>Video (coming soon)</strong></h4>
</li>
<li class="horizItem">
<a href="/zerograds">
<a href="https://colab.research.google.com/drive/1GNd6DdRGRHQjKyG3rIfJ9519lo9VuAq4?usp=sharing">
<img class="teaserbutton" src="/assets/images/colablogo.png"><br>
<h4><strong>Colab (coming soon)</strong></h4>
<h4><strong>Colab</strong></h4>
</a>
</li>
<li class="horizItem">
Expand Down Expand Up @@ -134,7 +134,7 @@ therefore can "focus on what matters" at the current optimization step. In simpl
<li>Compute the gradient of the MLP w.r.t. the optimization parameters, and use this gradient to perform an (Adam-) gradient descent step.</li>
</ol>

We provide a <a href="/zerograds">Colab Notebook</a> that shows a simple example and how ZeroGrads outperforms its competitors.
We provide a <a href="https://colab.research.google.com/drive/1GNd6DdRGRHQjKyG3rIfJ9519lo9VuAq4?usp=sharing">Colab Notebook</a> that shows a simple example and how ZeroGrads outperforms its competitors.

<br>
<br>
Expand Down Expand Up @@ -185,14 +185,14 @@ previously published gradient estimator <a href="/prdpt">PRDPT</a>, here denoted
We use a weighted mixture of MSE and KLD as the trainig loss and backpropagate this loss through the non-diff. spline
renderer (i.e., from the rendered spline to the VAE weights) using our proposed method ZeroGrads. The below digits
are samples from the latent space of the trained VAE, in a variety of styles, which are easily applicable in post-processing
thanks to the spline formulation.
thanks to the spline formulation.<br>
</p>

<div class="container">
<img class="containerimg" src="/assets/images/zerograds/vae.png">
<div class="caption"></div>
</div>

<br>
<br>

The magic that makes our approach scale to high dimensions is the surrogate's hysteresis, which reduces the gradient
Expand Down
2 changes: 1 addition & 1 deletion _site/feed.xml
View file
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
<?xml version="1.0" encoding="utf-8"?><feed xmlns="http://www.w3.org/2005/Atom" ><generator uri="https://jekyllrb.com/" version="3.9.2">Jekyll</generator><link href="http://localhost:4000/feed.xml" rel="self" type="application/atom+xml" /><link href="http://localhost:4000/" rel="alternate" type="text/html" /><updated>2024-05-13T11:04:51+01:00</updated><id>http://localhost:4000/feed.xml</id><title type="html">Michael Fischer</title><subtitle>Michael Fischer, PhD Student in AI and Computer Graphics at University College London (UCL).</subtitle><author><name>Michael Fischer</name></author><entry><title type="html">Welcome to Jekyll!</title><link href="http://localhost:4000/blog/welcome-to-jekyll/" rel="alternate" type="text/html" title="Welcome to Jekyll!" /><published>2019-04-18T20:34:30+01:00</published><updated>2019-04-18T20:34:30+01:00</updated><id>http://localhost:4000/blog/welcome-to-jekyll</id><content type="html" xml:base="http://localhost:4000/blog/welcome-to-jekyll/">&lt;p&gt;You’ll find this post in your &lt;code class=&quot;language-plaintext highlighter-rouge&quot;&gt;_posts&lt;/code&gt; directory. Go ahead and edit it and re-build the site to see your changes. You can rebuild the site in many different ways, but the most common way is to run &lt;code class=&quot;language-plaintext highlighter-rouge&quot;&gt;jekyll serve&lt;/code&gt;, which launches a web server and auto-regenerates your site when a file is updated.&lt;/p&gt;
<?xml version="1.0" encoding="utf-8"?><feed xmlns="http://www.w3.org/2005/Atom" ><generator uri="https://jekyllrb.com/" version="3.9.2">Jekyll</generator><link href="http://localhost:4000/feed.xml" rel="self" type="application/atom+xml" /><link href="http://localhost:4000/" rel="alternate" type="text/html" /><updated>2024-05-13T12:47:17+01:00</updated><id>http://localhost:4000/feed.xml</id><title type="html">Michael Fischer</title><subtitle>Michael Fischer, PhD Student in AI and Computer Graphics at University College London (UCL).</subtitle><author><name>Michael Fischer</name></author><entry><title type="html">Welcome to Jekyll!</title><link href="http://localhost:4000/blog/welcome-to-jekyll/" rel="alternate" type="text/html" title="Welcome to Jekyll!" /><published>2019-04-18T20:34:30+01:00</published><updated>2019-04-18T20:34:30+01:00</updated><id>http://localhost:4000/blog/welcome-to-jekyll</id><content type="html" xml:base="http://localhost:4000/blog/welcome-to-jekyll/">&lt;p&gt;You’ll find this post in your &lt;code class=&quot;language-plaintext highlighter-rouge&quot;&gt;_posts&lt;/code&gt; directory. Go ahead and edit it and re-build the site to see your changes. You can rebuild the site in many different ways, but the most common way is to run &lt;code class=&quot;language-plaintext highlighter-rouge&quot;&gt;jekyll serve&lt;/code&gt;, which launches a web server and auto-regenerates your site when a file is updated.&lt;/p&gt;

&lt;p&gt;To add new posts, simply add a file in the &lt;code class=&quot;language-plaintext highlighter-rouge&quot;&gt;_posts&lt;/code&gt; directory that follows the convention &lt;code class=&quot;language-plaintext highlighter-rouge&quot;&gt;YYYY-MM-DD-name-of-post.ext&lt;/code&gt; and includes the necessary front matter. Take a look at the source for this post to get an idea about how it works.&lt;/p&gt;

Expand Down
10 changes: 5 additions & 5 deletions _site/zerograds/index.html
View file
Original file line number Diff line number Diff line change
Expand Up @@ -400,9 +400,9 @@ <h4><strong>Paper</strong></h4>
<h4><strong>Video (coming soon)</strong></h4>
</li>
<li class="horizItem">
<a href="/zerograds">
<a href="https://colab.research.google.com/drive/1GNd6DdRGRHQjKyG3rIfJ9519lo9VuAq4?usp=sharing">
<img class="teaserbutton" src="/assets/images/colablogo.png" /><br />
<h4><strong>Colab (coming soon)</strong></h4>
<h4><strong>Colab</strong></h4>
</a>
</li>
<li class="horizItem">
Expand Down Expand Up @@ -464,7 +464,7 @@ <h4><strong>Supplemental</strong></h4>
<li>Compute the gradient of the MLP w.r.t. the optimization parameters, and use this gradient to perform an (Adam-) gradient descent step.</li>
</ol>

We provide a <a href="/zerograds">Colab Notebook</a> that shows a simple example and how ZeroGrads outperforms its competitors.
We provide a <a href="https://colab.research.google.com/drive/1GNd6DdRGRHQjKyG3rIfJ9519lo9VuAq4?usp=sharing">Colab Notebook</a> that shows a simple example and how ZeroGrads outperforms its competitors.

<br />
<br />
Expand Down Expand Up @@ -515,14 +515,14 @@ <h4><strong>Supplemental</strong></h4>
We use a weighted mixture of MSE and KLD as the trainig loss and backpropagate this loss through the non-diff. spline
renderer (i.e., from the rendered spline to the VAE weights) using our proposed method ZeroGrads. The below digits
are samples from the latent space of the trained VAE, in a variety of styles, which are easily applicable in post-processing
thanks to the spline formulation.
thanks to the spline formulation.<br />
</p>

<div class="container">
<img class="containerimg" src="/assets/images/zerograds/vae.png" />
<div class="caption"></div>
</div>

<br />
<br />

The magic that makes our approach scale to high dimensions is the surrogate's hysteresis, which reduces the gradient
Expand Down

0 comments on commit e5944a6

Please sign in to comment.