Minor edits to fit the Acknowledgement section

srds22/conclu.tex

@@ -43,10 +43,7 @@ nodes, non-uniform latency between pairs of nodes) and dynamically constructing
 % be constructed in a decentralized way with
 % PeerSampling~\cite{jelasity2007gossip}. This will be investigated in future work, 
 % using these results as a design foundation.
-More general types of data heterogeneity could be tackled beyond the
-important case
-of
-label distribution skew on which we focused in this paper. An important
+More general types of data heterogeneity could be tackled: An important
 example is
 covariate shift or feature distribution skew \cite{kairouz2019advances}, for
 which local density estimates could be used as basis to construct cliques that

srds22/d-cliques.tex

@@ -50,13 +50,13 @@ impractical.
 
 \begin{figure}[t]
      \centering
-     \begin{subfigure}[b]{0.18\textwidth}
+     \begin{subfigure}[b]{0.16\textwidth}
          \centering
          \includegraphics[width=\textwidth]{figures/grid-iid-neighbourhood}
 \caption{\label{fig:grid-iid-neighbourhood} Homogeneous data}
      \end{subfigure}
      \hspace*{.5cm}
-     \begin{subfigure}[b]{0.18\textwidth}
+     \begin{subfigure}[b]{0.16\textwidth}
          \centering
          \includegraphics[width=\textwidth]{figures/grid-non-iid-neighbourhood}
 \caption{\label{fig:grid-non-iid-neighbourhood}  Heterogeneous data}
@@ -316,7 +316,7 @@ log-linear in $n$.
 
 \begin{figure}[t]
     \centering
-    \includegraphics[width=0.20\textwidth]{figures/fully-connected-cliques}
+    \includegraphics[width=0.18\textwidth]{figures/fully-connected-cliques}
     \caption{\label{fig:d-cliques-figure} D-Cliques with $n=100$, $M=10$ and a
 fully connected inter-clique topology on a problem with 1 class/node.}
 \end{figure}
@@ -414,7 +414,7 @@ of the local models across nodes.
 
 \begin{figure}[t]
          \centering
-         \includegraphics[width=0.3\textwidth]{figures/connected-cliques-bias}
+         \includegraphics[width=0.28\textwidth]{figures/connected-cliques-bias}
 \caption{\label{fig:connected-cliques-bias} Illustrating the bias induced by
 inter-clique connections (see main text for details).}
 \end{figure}

srds22/exp.tex

@@ -618,7 +618,7 @@ slower.
 %python $TOOLS/analyze/filter.py skews --topology:name d-cliques/greedy-swap | python $TOOLS/plot/skew/convergence.py --max-steps 400 --labels 'MNIST (unbalanced classes)' 'CIFAR10 (balanced classes)' --linewidth 2.5 --save-figure ../mlsys2022style/figures/skew-convergence-speed-2-shards.png --linestyles 'solid' 'dashed'
 \begin{figure}[htbp]
     \centering
-    \includegraphics[width=0.25\textwidth]{figures/skew-convergence-speed-2-shards}
+    \includegraphics[width=0.23\textwidth]{figures/skew-convergence-speed-2-shards}
     \caption{\label{fig:skew-convergence-speed-2-shards} Skew decrease during clique construction of 10 cliques of 10 heterogeneous nodes (100 nodes). Bold line is the average over 100 experiments. Thin lines are respectively the minimum and maximum over all experiments. In wall-clock time, 1000 steps take less than 6 seconds in Python 3.7 on a MacBook Pro 2020.}
 \end{figure}
 
@@ -659,5 +659,4 @@ is therefore critical to convergence speed.
 In an extended version of this paper~\cite{dcliques-arxiv}, we replicate experimental
 results on an extreme case of label distribution skew where each node only has
 examples of a single class. These results consistently show that our
-approach remains effective even for extremely skewed label distributions
-across nodes.
+approach remains effective even for extremely skewed label distributions.

srds22/main.tex

@@ -144,6 +144,11 @@ Heterogeneous Data, Stochastic Gradient Descent
 \input{related_work}
 \input{conclu}
 
+\section*{Acknowledgements}
+
+This work was supported in part by the French National Research Agency
+(ANR) through grant ANR-20-CE23-0015 (Project PRIDE) and ANR-16-CE23-0016-01 (Project PAMELA).
+
 \bibliographystyle{IEEEtran}
 \bibliography{IEEEabrv,main.bib}
 

srds22/related_work.tex

@@ -10,9 +10,9 @@ algorithms.
 \paragraph{Dealing with heterogeneity in server-based FL}
 Data heterogeneity is not much of an issue in server-based FL if
 clients send their parameters to the server after each gradient update.
-Problems arise when one seeks to reduce
-the number of communication rounds by allowing each participant to perform
-multiple local updates, as in the popular FedAvg algorithm 
+Problems arise when participants perform
+multiple local updates to reduce
+the number of communication rounds, as in the popular FedAvg algorithm 
 \cite{mcmahan2016communication}. Indeed, data heterogeneity can prevent
 such algorithms from
 converging to a good solution \cite{quagmire,scaffold}. This led to the design