add sentence on analyses in conclu

icdcs22/appendix_diverse_only.tex

@@ -51,7 +51,7 @@ CIFAR10, the clustering provided by D-Cliques, in combination with Clique
 Averaging, provides faster convergence and lower variance than diverse
 neighborhoods.
 
-In practice, the difficulty of obtaining a random graph in which each node has a neighborhood with similar skew as D-Cliques, while not using more edges, actually makes the construction with a decentralized algorithm harder. Indeed, there are more constraints in selecting potential candidates for swapping.
+In practice, the difficulty of obtaining a random graph in which each node has a neighborhood with similar skew as D-Cliques, while not using more edges, makes the construction with a decentralized algorithm harder. Indeed, there are more constraints in selecting potential candidates for swapping.
 
 \begin{figure}[t]
      \centering     

icdcs22/conclu.tex

@@ -23,6 +23,10 @@ results and that a small-world inter-clique topology with only $O(n \log n)$
 edges achieves a very good compromise between
 convergence speed and scalability with the number of nodes.
 
+More generally, our empirical results call for refined convergence analyses of
+decentralized FL that precisely capture the interplay between data
+heterogeneity and network topology.
+
 The ideas of D-Cliques can be useful to
 implement fully decentralized
 learning in a