Merge pull request #2 from SonyResearch/hojoonlee/update_project_page

Hojoonlee/update project page
SonyResearch · Oct 15, 2024 · a25db74 · a25db74
2 parents be01681 + b0b8a18
commit a25db74
Show file tree

Hide file tree

Showing 4 changed files with 55 additions and 12 deletions.
diff --git a/README.md b/README.md
@@ -1,6 +1,41 @@
-# SimBa: Simplicity Bias for Scaling Up Parameters in Deep Reinforcement Learning
+# SimBa: Simplicity Bias for Scaling Up Parameters in Deep RL
 
-This is a repository of an official implementation of Simba: Simplicity Bias for Scaling Up Parameters in Deep Reinforcement Learning.
+This is a repository of an official implementation of 
+
+<i>Simba: Simplicity Bias for Scaling Up Parameters in Deep Reinforcement Learning </i> by
+
+<a href="https://joonleesky.github.io">Hojoon Lee</a>,
+<a href="https://godnpeter.github.io">Dongyoon Hwang</a>,
+<a href="https://i-am-proto.github.io">Donghu Kim</a>,
+<a href="https://mynsng.github.io">Hyunseung Kim</a>,
+<a href="https://taijunjet.com">Jun Jet Tai</a>,
+<a href="https://kausubbu.github.io">Kaushik Subramanian</a>, 
+
+<a href="https://www.pwurman.org">Peter R. Wurman</a>,
+<a href="https://sites.google.com/site/jaegulchoo">Jaegul Choo</a>,
+<a href="https://www.cs.utexas.edu/~pstone/">Peter Stone</a>,
+<a href="https://takuseno.github.io/">Takuma Seno</a>.
+
+
+[[Website]](https://sonyresearch.github.io/simba) [[Paper]](https://arxiv.org/abs/2410.09754)
+
+## Overview
+
+### TL;DR
+
+Stop worrying about algorithms, just change the network architecture to SimBa.
+
+### Method
+
+SimBa is a network architecture designed for RL that avoids overfitting by embedding simplicity bias.
+
+<img src="docs/images/simba_architecture.png" alt="Image description" width="800">
+
+### Results
+
+When integrated SimBA with Soft Actor Critic (SAC), it matches the performance of state-of-the-art RL algorithms.
+
+<img src="docs/images/overview.png" alt="Image description" width="800">
 
 
 ## Getting strated
@@ -105,8 +140,9 @@ If you find our work useful, please consider citing our paper as follows:
 
 ```
 @article{lee2024simba,
-  title={Simba: Simplicity Bias for Scaling Up Parameters in Deep Reinforcement Learning}, 
-  author={Hojoon Lee and Dongyoon Hwang and Donghu Kim and Hyunseung Kim and Jun Jet Tai and Kaushik Subramanian and Peter R.Wurman and Jaegul Choo and Peter Stone and Takuma Seno},
+  title={SimBa: Simplicity Bias for Scaling Up Parameters in Deep Reinforcement Learning}, 
+  author={Hojoon Lee and Dongyoon Hwang and Donghu Kim and Hyunseung Kim and Jun Jet Tai and Kaushik Subramanian and Peter R. Wurman and Jaegul Choo and Peter Stone and Takuma Seno},
+  journal={arXiv preprint arXiv:2410.09754},
   year={2024}
 }
 ```
diff --git a/deps/environment.yaml b/deps/environment.yaml
@@ -31,3 +31,5 @@ dependencies:
     - termcolor==2.4.0
     - tqdm==4.66.1
     - wandb==0.16.6
+    - moviepy==1.0.3
+    - imageio==2.33.1
diff --git a/deps/requirements.txt b/deps/requirements.txt
@@ -20,3 +20,5 @@ tensorflow-probability==0.24.0
 termcolor==2.4.0
 tqdm==4.66.1
 wandb==0.16.6
+moviepy==1.0.3
+imageio==2.33.1
diff --git a/docs/index.html b/docs/index.html
@@ -17,7 +17,7 @@
 </head>
 <div class="header" id="top">
 	<h1><span class="bold simba">SimBa </span><br/>Simplicity Bias for Scaling Parameters in Deep Reinforcement Learning</h1>
-	<h3><a class="bold default-color">Under review</span><br/></h3>
+	<h3><a class="bold default-color">Preprint</span><br/></h3>
 	<table class="authors">
 		<tbody>
 			<tr>
@@ -49,7 +49,7 @@ <h4>
 		</tbody>
 	</table>
 	<div class="links">
-		<a href="https://arxiv.org" class="btn"><i class="fa">&#xf1c1;</i>&ensp;Paper</a><a href="https://github.com/joonleesky/scale_rl" class="btn"><i class="fa fa-github"></i>&ensp;Code</a>
+		<a href="http://arxiv.org/abs/2410.09754" class="btn"><i class="fa">&#xf1c1;</i>&ensp;Paper</a><a href="https://github.com/SonyResearch/simba" class="btn"><i class="fa fa-github"></i>&ensp;Code</a>
 	</div>
 </div>
 <div class="content">
@@ -313,7 +313,7 @@ <h2>Paper</h2>
 		<span class="italic">Hojoon Lee&ast;, Dongyoon Hwang&ast;, Donghu Kim,<br/>
 							Hyunseung Kim, Jun Jet Tai, Kaushik Subramanian, Peter R. Wurman,<br/>
 							Jaegul Choo, Peter Stone, Takuma Seno</span><br/><br/>
-		<a href="https://arxiv.org">arXiv preprint</a><br/><br/>
+		<a href="https://arxiv.org/abs/2410.09754">arXiv preprint</a><br/><br/>
 		<div class="page" style="background-image: url(thumbnails/0.png);"></div>
 		<div class="page" style="background-image: url(thumbnails/1.png);"></div>
 		<div class="page" style="background-image: url(thumbnails/2.png);"></div>
@@ -324,9 +324,6 @@ <h2>Paper</h2>
 		<div class="page" style="background-image: url(thumbnails/7.png);"></div>
 		<div class="page" style="background-image: url(thumbnails/8.png);"></div>
 		<div class="page" style="background-image: url(thumbnails/9.png);"></div>
-		<div style="margin: auto; margin-top: 32px;">
-			<a href="https://arxiv.org/abs/2310.16828">View on arXiv</a>
-		</div>
 	</div>
 	<div class="hr"></div>
 	<div style="padding-bottom: 64px; text-align: center;">
@@ -335,8 +332,14 @@ <h2>Citation</h2>
 			If you find our work useful, please consider citing the paper as follows:
 		</p>
 		<div id="bibtex-text" class="bibtexsection" onClick="window.getSelection().selectAllChildren(document.getElementById('bibtex-text'));">
-
-</div>
+@article{lee2024simba,
+	title={SimBa: Simplicity Bias for Scaling Up Parameters in Deep Reinforcement Learning}, 
+	author={Hojoon Lee and Dongyoon Hwang and Donghu Kim and Hyunseung Kim and Jun Jet Tai and 
+		Kaushik Subramanian and Peter R. Wurman and Jaegul Choo and Peter Stone and Takuma Seno},
+	journal={arXiv preprint arXiv:2410.09754},
+	year={2024}
+}
+		</div>
 	</div>
 </div>
 <footer>