Unify slides

slides.py

@@ -44,18 +44,18 @@ def sort_paths(plist):
         number1 = int(''.join(filter(str.isdigit, nxt_element.name)))
         number2 = int(''.join(filter(str.isdigit, plist[j].name)))
 
-        if number1 >= 10:
+        if number1 > 10 and number1 <= 90:
             number1 /= 10
 
-        if number2 >= 10:
+        if number2 > 10 and number2 <= 90:
             number2 /= 10
 
         while number2 > number1  and j >= 0:
             plist[j+1] = plist[j]
             j = j - 1
             number2 = int(''.join(filter(str.isdigit, plist[j].name)))
 
-            if number2 >= 10:
+            if number2 > 10 and number2 <= 90:
                 number2 /= 10
         plist[j+1] = nxt_element    
 

w09_nas/t07_oneshot_nas.tex

@@ -28,10 +28,10 @@
 		\item The first type of one-shot models: \alert{convolutional neural fabrics} %/ \alert{lattice} 
 
 	\centering
-	\only<2-5>{\includegraphics[width=0.7\textwidth]{images/conv_fabric_1.png}\\}
-	\only<6>{\hspace*{-0.91cm}\vspace*{0.3cm}\includegraphics[width=1.02\textwidth]{images/conv_fabric_3.png}\vspace*{-0.5cm}}
+	\only<2-5>{\includegraphics[width=0.7\textwidth]{images/conv_fabric_1.jpg}\\}
+	\only<6>{\hspace*{-0.91cm}\vspace*{0.3cm}\includegraphics[width=1.02\textwidth]{images/conv_fabric_3.jpg}\vspace*{-0.5cm}}
 
-	\only<2>{\bigskip\bigskip\includegraphics[width=0.7\textwidth]{images/conv_fabric_2.png}}
+	\only<2>{\bigskip\bigskip\includegraphics[width=0.7\textwidth]{images/conv_fabric_2.jpg}}
 \pause
 	\medskip
 		\myit{
@@ -321,20 +321,23 @@
 	\end{itemize}
 
 	\onslide<5>{
-    	\begin{minipage}{0.4\textwidth}
+	\begin{columns}
+	    \column{0.01\textwidth}
+    	\column{0.4\textwidth}
         	\begin{itemize}
 				\footnotesize
 				\item \textbf{Pitfall:} the correlation between architectures evaluated with the one-shot weights and retrained from scratch (stand-alone models) should be high
 				\item If not, \textbf{selecting the best architecture based on the one-shot weights} is sub-optimal.
 			\end{itemize}
-	    \end{minipage}
-	    \hspace{1cm}
-    	\begin{minipage}{0.5\textwidth}
+	    \column{0.5\textwidth}
 \begin{center}
+\vspace*{-0.3cm}
 	        \includegraphics[width=.7\textwidth]{images/bender_correlation_1.png}\\
 	        \scriptsize{From \smaller{\lit{\href{http://proceedings.mlr.press/v80/bender18a/bender18a.pdf}{Bender et al. 2018}}}}
 \end{center}
-    	\end{minipage}
+	    \column{0.09\textwidth}
+
+    	\end{columns}
 	}
 }
 %---------------------------------------------------------------

w09_nas/t09_NASlib.tex

@@ -9,7 +9,7 @@
 \begin{document}
 \maketitle
 %----------------------------------------------------------------------
-\myframe{Motivation for NASLib \litw{Zela et al., 2020}}{
+\myframe{Motivation for NASLib \litw{Zela et al. 2020}}{
 
 	\alert{NASLib} is a \alert{framework} for easily implementing different NAS methods, aiming to:
 \medskip		
@@ -132,13 +132,13 @@
 \myframe{Tabular benchmarks for one-shot NAS}{
 	\centering
 	\myit{
-		\item NAS-Bench-101 \lit{\href{https://arxiv.org/abs/1902.09635}{Ying et al. 2019}} is not directly compatible with one-shot NAS methods 
+		\item NAS-Bench-101 \lit{\href{https://arxiv.org/pdf/1902.09635.pdf}{Ying et al. 2019}} is not directly compatible with one-shot NAS methods 
 		\myit{
 			\item Mainly due to the constraint of at most 9 edges in the cell
 		}
 		\bigskip
 		\pause
-		\item \alert{NAS-Bench-1Shot1} \lit{\href{https://openreview.net/forum?id=SJx9ngStPH}{Zela et al. 2020}}
+		\item \alert{NAS-Bench-1Shot1} \lit{\href{https://openreview.net/pdf?id=SJx9ngStPH}{Zela et al. 2020}}
 		\myit{
 				\item[-] \alert{3 sub-spaces of NAS-Bench-101} that are compatible with one-shot methods
 				\myit{
@@ -148,7 +148,7 @@
 			}
 		\bigskip
 		\pause
-		\item \alert{NAS-Bench-201} \lit{\href{https://openreview.net/forum?id=HJxyZkBKDr}{Dong and Yang. 2020}}
+		\item \alert{NAS-Bench-201} \lit{\href{https://openreview.net/pdf?id=HJxyZkBKDr}{Dong and Yang. 2020}}
 		\myit{
 %				\item[-] Same graph representation as in DARTS
 				\item[-] Much smaller than NAS-Bench-101 and largest NAS-Bench-1Shot1 subspace
@@ -167,7 +167,7 @@
 		\myit{
 			\item<1-> NAS-Bench-201 is already integrated in NASLib and we can run any one-shot optimizer on it
 \medskip
-			\item<2-> We can also combine random perturbations \lit{\href{https://arxiv.org/pdf/2002.05283.pdf}{Chen and Hsieh, 2020}} with any one-shot optimizer
+			\item<2-> We can also combine random perturbations \lit{\href{https://arxiv.org/pdf/2002.05283.pdf}{Chen and Hsieh. 2020}} with any one-shot optimizer
 \medskip
 			\item<3-> We can also evaluate black-box optimizers cheaply with a tabular benchmark
 		}	

w09_nas/t10_practical_recommendations.tex

@@ -62,14 +62,14 @@
 			\item[-] If the compute budget suffices, \alert{optimize them jointly}, e.g., using BOHB
 			\myit{
 				\item[+] Auto-PyTorch Tabular \lit{\href{https://github.com/automl/Auto-PyTorch}{Zimmer,\,Lindauer\,\& Hutter, 2020}}
-				\item[+] Auto-RL \lit{\href{https://arxiv.org/abs/1812.11951}{Runge et al, 2019}}
+				\item[+] Auto-RL \lit{\href{https://arxiv.org/pdf/1812.11951.pdf}{Runge et al. 2019}}
 			}
 \medskip
 \pause
 			\item[-] Else
 			\myit{
 				\item[+] If you have decent hyperparameters:\\
-			\alert{run NAS, followed by HPO for fine-tuning} \lit{\href{https://arxiv.org/abs/1905.07443}{Saikat et al, 2019}}
+			\alert{run NAS, followed by HPO for fine-tuning} \lit{\href{https://arxiv.org/pdf/1905.07443.pdf}{Saikat et al. 2019}}
 \smallskip
 \pause
 				\item[+] If you don't have decent hyperparameters: \alert{first run HPO} to get competitive	
@@ -110,18 +110,18 @@
 
 	\column{0.29\textwidth}	
 \onslide<1->{
-	\includegraphics[width=\linewidth]{images/AutoPytorch-ShapedResNet.png}\\
+	\includegraphics[width=\linewidth]{images/AutoPytorch-ShapedResNet.jpg}\\
 }
 	\column{0.42\textwidth}	
 \vspace*{-0.7cm}
 \onslide<1->{
 	\begin{center}
-	\includegraphics[width=0.9\linewidth]{images/AutoPytorch-ConfigSpace.png}
+	\includegraphics[width=0.9\linewidth]{images/AutoPytorch-ConfigSpace.jpg}
 	\end{center}
 }
 \onslide<5->{
 	\vspace*{-0.3cm}\hspace*{-1.1cm}
-	\includegraphics[width=1.1\linewidth]{images/AutoPytorch-performance.png}
+	\includegraphics[width=1.1\linewidth]{images/AutoPytorch-performance.jpg}
 %	\end{center}
 }
 
@@ -158,7 +158,7 @@
 	}
 	\onslide<4->{
 		\myit{
-			\item Both NAS and HPO improved the state of the art \lit{\href{https://arxiv.org/abs/1905.07443}{Saikat et al, 2019}}:
+			\item Both NAS and HPO improved the state of the art \lit{\href{https://arxiv.org/pdf/1905.07443.pdf}{Saikat et al. 2019}}:
 			\myit{
 				\item End-point-error (EPE) on Sintel dataset: \alert{2.36 $\rightarrow$ 2.14 (by DARTS)}
 				\item Subsequent HPO: \alert{2.14 $\rightarrow$ 1.94 (by BOHB)}