Updated experiment parameters

main.tex

@@ -131,8 +131,10 @@ Because the data distribution within each clique is representative of the global
 
 As a summary, we make the following contributions:
 \begin{itemize}
+  \item significant impact of topology on non-iid data
   \item we propose the D-Cliques topology to remove the impact of non-IID data on convergence speed, similar to a fully-connected topology, with a reduced number of edges and required messages
-  \item we show how to leverage D-Cliques to implement momentum in a distributed non-IID setting, which would otherwise be detrimental to the convergence speed of convolutional networks 
+  \item we show how to leverage D-Cliques to implement momentum in a distributed non-IID setting, which would otherwise be detrimental to the convergence speed of convolutional networks
+  \item scale (>16 noeuds)
 \end{itemize}
 
 The rest of the paper is organized as such. \dots
@@ -276,6 +278,8 @@ We solve this problem by decoupling the gradient averaging from the weight avera
 \caption{\label{fig:d-cliques-mnist-linear} D-Cliques with Linear Model on MNIST.}
 \end{figure}
 
+With and without clique averaging.
+
 TODO: Update figure with actual Clique-Ring results
 
 \subsection{CIFAR10 and Convolutional Model}
@@ -330,13 +334,23 @@ Similar number of maximum hops but no or less clustering than D-Cliques (and no
 % To regenerate the figure, from directory results/scaling
 % python ../../../learn-topology/tools/plot_convergence.py 10/mnist/fully-connected-cliques/all/2021-03-10-14:40:35-CET ../mnist/fully-connected-cliques/all/2021-03-10-10:19:44-CET 1000/mnist/fully-connected-cliques/all/2021-03-10-16:44:35-CET --labels '10 nodes bsz=128' '100 nodes bsz=128' '1000 nodes bsz=128 (45)' --legend 'lower right' --yaxis test-accuracy --save-figure ../../figures/d-cliques-mnist-scaling-fully-connected.png --ymin 80 --add-min-max
      \begin{figure}[htbp]
+         \centering
+         \begin{subfigure}[b]{\textwidth}
          \centering
          \includegraphics[width=0.7\textwidth]{figures/d-cliques-mnist-scaling-fully-connected}
-\caption{\label{fig:d-cliques-mnist-scaling-fully-connected} Scaling Behaviour of Fully-Connected D-Clique}
+         \caption{Constant Batch-Size}
+     \end{subfigure}
+      \begin{subfigure}[b]{\textwidth}
+         \centering
+         \caption{Constant Nb Updates per Epoch}
+     \end{subfigure}
+     \caption{\label{fig:d-cliques-mnist-scaling-fully-connected} Scaling Behaviour of Fully-Connected D-Clique}
      \end{figure}
 
 Show scaling effect for 10, 100, 1000 nodes (with decreasing sample sizes) for Clique Ring, Hierarchical, Fully-Connected.
 
+(Smallworld?)
+
 Robustness to not having fully-connected cliques (static and dynamic subsets).
 
 \section{Related Work}

results/scaling/10/mnist/fully-connected-cliques/experiments.sh

 #!/usr/bin/env bash
 TOOLS=../../../../../../learn-topology/tools; CWD="$(pwd)"; cd $TOOLS
 BSZS='
-    128
+    1280
     '
 LRS='
     0.1

results/scaling/1000/mnist/fully-connected-cliques/experiments.sh

 #!/usr/bin/env bash
 TOOLS=../../../../../../learn-topology/tools; CWD="$(pwd)"; cd $TOOLS
 BSZS='
-    128
+    13
     '
 LRS='
     0.1