Skip to content
Snippets Groups Projects
Commit 1ca8944e authored by Erick Lavoie's avatar Erick Lavoie
Browse files

Added some initial experiment results

parent a780ff0a
No related branches found
No related tags found
No related merge requests found
Hide whitespace changes
Inline Side-by-side
Showing
with 224 additions and 144 deletions
mlsys2022style/appendix.tex
+ 54
− 0
View file @ 1ca8944e
...@@ -359,5 +359,59 @@ using the corresponding input functions. ...@@ -359,5 +359,59 @@ using the corresponding input functions.
\subsection{Evaluation} \subsection{Evaluation}
\label{section:ideal-cliques-evaluation} \label{section:ideal-cliques-evaluation}
\subsubsection{Convergence Speed of D-Cliques Compared to Fully-Connected}
% From directory 'results-v2':
% MNIST
% python $TOOLS/analyze/filter.py all --dataset:name mnist --topology:name fully-connected d-cliques/ideal --nodes:name max-local-skew --meta:seed 1 --nodes:nb-nodes 100 | python $TOOLS/analyze/diff.py
% python $TOOLS/analyze/diff.py --rundirs all/2021-09-28-12:39:00-CEST-labostrex117 all/2021-09-28-23:18:40-CEST-labostrex118 --pass-through | python $TOOLS/plot/convergence.py --add-min-max --ymin 80 --ymax 92.5 --yaxis test-accuracy --labels 'fully-connected' 'd-cliques (fc) w/ cliq-avg' --save-figure ../mlsys2022style/figures/convergence-speed-mnist-dc-fc-vs-fc-1-class-per-node.png --linestyles 'solid' 'dashed'
% CIFAR10
% python $TOOLS/analyze/filter.py all --dataset:name cifar10 --topology:name fully-connected d-cliques/ideal --nodes:name max-local-skew --meta:seed 1 --nodes:nb-nodes 100 | python $TOOLS/analyze/diff.py
% python $TOOLS/analyze/diff.py --rundirs all/2021-10-03-16:09:21-CEST-labostrex112 all/2021-10-03-19:45:14-CEST-labostrex118 --pass-through | python $TOOLS/plot/convergence.py --add-min-max --ymin 0 --ymax 100 --yaxis test-accuracy --labels 'fully-connected w/ momentum' 'd-cliques (fc) w/ cliq-avg and momentum' --save-figure ../mlsys2022style/figures/convergence-speed-cifar10-dc-fc-vs-fc-1-class-per-node.png --linestyles 'solid' 'dashed' --legend 'lower right'
\begin{figure}[htbp]
\centering
\begin{subfigure}[b]{0.35\textwidth}
\centering
\includegraphics[width=\textwidth]{figures/convergence-speed-mnist-dc-fc-vs-fc-1-class-per-node}
\caption{\label{fig:convergence-speed-mnist-dc-fc-vs-fc-1-class-per-node} MNIST}
\end{subfigure}
\hfill
\begin{subfigure}[b]{0.35\textwidth}
\centering
\includegraphics[width=\textwidth]{figures/convergence-speed-cifar10-dc-fc-vs-fc-1-class-per-node}
\caption{\label{fig:convergence-speed-cifar10-dc-fc-vs-fc-1-class-per-node} CIFAR10}
\end{subfigure}
\caption{\label{fig:convergence-speed-dc-vs-fc-1-class-per-node} Convergence Speed of D-Cliques Compared to Fully-Connected on 100 Nodes (1 class/node).}
\end{figure}
\subsubsection{Effect of Removing Intra-clique Edges}
% From directory 'results-v2':
% MNIST
% python $TOOLS/analyze/filter.py all --dataset:name mnist --topology:name d-cliques/ideal --nodes:name max-local-skew --meta:seed 1 --nodes:nb-nodes 100 | python $TOOLS/analyze/diff.py
% w/o Clique Gradient
% python $TOOLS/analyze/diff.py --rundirs all/2021-09-29-03:52:47-CEST-labostrex118 all/2021-10-02-21:26:18-CEST-labostrex113 all/2021-10-03-06:33:52-CEST-labostrex113 --pass-through | python $TOOLS/plot/convergence.py --add-min-max --ymin 89 --ymax 92.5 --yaxis test-accuracy --labels 'full intra-connectivity' '-1 edge/clique' '-5 edges/clique' --save-figure ../mlsys2022style/figures/d-cliques-ideal-wo-clique-avg-impact-of-edge-removal.png --linestyles 'solid' 'dashed' 'dotted' --font-size 18
% w/ Clique Gradient
% python $TOOLS/analyze/diff.py --rundirs all/2021-09-28-23:18:40-CEST-labostrex118 all/2021-10-02-16:50:53-CEST-labostrex113 all/2021-10-03-02:00:23-CEST-labostrex113 --pass-through | python $TOOLS/plot/convergence.py --add-min-max --ymin 89 --ymax 92.5 --yaxis test-accuracy --labels 'full intra-connectivity' '-1 edge/clique' '-5 edges/clique' --save-figure ../mlsys2022style/figures/d-cliques-ideal-w-clique-avg-impact-of-edge-removal.png --linestyles 'solid' 'dashed' 'dotted' --font-size 18
\begin{figure}[t]
\centering
\begin{subfigure}[htbp]{0.23\textwidth}
\centering
\includegraphics[width=\textwidth]{figures/d-cliques-ideal-wo-clique-avg-impact-of-edge-removal}
\caption{\label{fig:d-cliques-ideal-wo-clique-avg-impact-of-edge-removal} Without Clique Averaging }
\end{subfigure}
\hfill
\begin{subfigure}[htbp]{0.23\textwidth}
\centering
\includegraphics[width=\textwidth]{figures/d-cliques-ideal-w-clique-avg-impact-of-edge-removal}
\caption{\label{fig:d-cliques-ideal-w-clique-avg-impact-of-edge-removal} With Clique Averaging}
\end{subfigure}
\caption{\label{fig:d-cliques-ideal-intra-connectivity} Impact of Intra-clique Connectivity Failures on MNIST (1 class/node). Y axis starts at 89.}
\end{figure}
\ No newline at end of file
mlsys2022style/d-cliques.tex
+ 7
− 0
View file @ 1ca8944e
...@@ -71,6 +71,13 @@ edge with all other cliques, see Figure~\ref{fig:d-cliques-figure} for the ...@@ -71,6 +71,13 @@ edge with all other cliques, see Figure~\ref{fig:d-cliques-figure} for the
corresponding D-Cliques network in the case of $n=100$ nodes and $L=10$ corresponding D-Cliques network in the case of $n=100$ nodes and $L=10$
classes. We will explore sparser inter-clique topologies in Section~\ref{section:interclique-topologies}. classes. We will explore sparser inter-clique topologies in Section~\ref{section:interclique-topologies}.
\begin{figure}[htbp]
\centering
\includegraphics[width=0.20\textwidth]{../figures/fully-connected-cliques}
\caption{\label{fig:d-cliques-figure} D-Cliques (fully-connected
cliques) example with 1 class/node.}
\end{figure}
We construct D-Cliques by initializing cliques at random, using at most $M$ We construct D-Cliques by initializing cliques at random, using at most $M$
nodes to limit the intra-clique communication costs, then we nodes to limit the intra-clique communication costs, then we
swap nodes between pairs of cliques chosen at random such that the swap swap nodes between pairs of cliques chosen at random such that the swap
......
mlsys2022style/exp.tex
+ 163
− 144
View file @ 1ca8944e
...@@ -77,126 +77,152 @@ mini-batch size, both approaches are equivalent. ...@@ -77,126 +77,152 @@ mini-batch size, both approaches are equivalent.
\subsection{D-Cliques match the Convergence Speed of Fully-Connected with a Fraction of the Edges} \subsection{D-Cliques match the Convergence Speed of Fully-Connected with a Fraction of the Edges}
\begin{figure}[t] % From directory 'results-v2':
\centering % MNIST
% python $TOOLS/analyze/filter.py all --dataset:name mnist --topology:name fully-connected d-cliques/greedy-swap --nodes:name 2-shards-uneq-classes --meta:seed 1 --nodes:nb-nodes 100 | python $TOOLS/analyze/diff.py
\begin{subfigure}[b]{0.20\textwidth} % python $TOOLS/analyze/diff.py --rundirs all/2021-09-28-23:16:47-CEST-labostrex117 all/2021-09-28-23:18:49-CEST-labostrex119 --pass-through | python $TOOLS/plot/convergence.py --add-min-max --ymin 80 --ymax 92.5 --yaxis test-accuracy --labels 'fully-connected' 'd-cliques (fc) w/ cliq-avg' --save-figure ../mlsys2022style/figures/convergence-speed-mnist-dc-fc-vs-fc-2-shards-per-node.png --linestyles 'solid' 'dashed' --font-size 18
% CIFAR10
% python $TOOLS/analyze/filter.py all --dataset:name cifar10 --topology:name fully-connected d-cliques/greedy-swap --nodes:name 2-shards-eq-classes --meta:seed 1 --nodes:nb-nodes 100 | python $TOOLS/analyze/diff.py
% python $TOOLS/analyze/diff.py --rundirs all/2021-10-02-18:58:22-CEST-labostrex114 all/2021-10-03-19:53:21-CEST-labostrex117 --pass-through | python $TOOLS/plot/convergence.py --add-min-max --ymin 0 --ymax 100 --yaxis test-accuracy --labels 'fully-connected w/ mom.' 'd-cliques (fc) w/ c-avg, mom.' --save-figure ../mlsys2022style/figures/convergence-speed-cifar10-dc-fc-vs-fc-2-shards-per-node.png --linestyles 'solid' 'dashed' --legend 'lower right' --font-size 18
\begin{figure}[htbp]
\centering
\begin{subfigure}[b]{0.23\textwidth}
\centering \centering
\includegraphics[width=\textwidth]{../figures/fully-connected-cliques} \includegraphics[width=\textwidth]{figures/convergence-speed-mnist-dc-fc-vs-fc-2-shards-per-node}
\caption{\label{fig:d-cliques-figure} D-Cliques (fully-connected \caption{\label{fig:convergence-speed-mnist-dc-fc-vs-fc-2-shards-per-node} MNIST}
cliques)}
\end{subfigure} \end{subfigure}
\hfill \hfill
% To regenerate figure, from results/mnist \begin{subfigure}[b]{0.23\textwidth}
% python ../../../../Software/non-iid-topology-simulator/tools/plot_convergence.py fully-connected/all/2021-03-10-09:25:19-CET no-init-no-clique-avg/fully-connected-cliques/all/2021-03-12-11:12:49-CET --add-min-max --yaxis test-accuracy --ymin 80 --ymax 92.5 --labels '100 nodes non-IID fully-connected' '100 nodes non-IID d-cliques' --save-figure ../../figures/d-cliques-mnist-vs-fully-connected.png --legend 'lower right' --font-size 16 --linestyles 'solid' 'dashed'
\begin{subfigure}[b]{0.26\textwidth}
\centering \centering
\includegraphics[width=\textwidth]{../figures/d-cliques-mnist-vs-fully-connected.png} \includegraphics[width=\textwidth]{figures/convergence-speed-cifar10-dc-fc-vs-fc-2-shards-per-node}
\caption{\label{fig:d-cliques-example-convergence-speed} Convergence Speed \caption{\label{fig:convergence-speed-cifar10-dc-fc-vs-fc-2-shards-per-node} CIFAR10}
on MNIST}
\end{subfigure} \end{subfigure}
\caption{\label{fig:convergence-speed-dc-vs-fc-2-shards-per-node} Convergence Speed of D-Cliques constructed with Greedy Swap Compared to Fully-Connected on 100 Nodes (2 shards/node).}
\caption{\label{fig:d-cliques-example} D-Cliques topology and convergence
speed on MNIST.}
\end{figure} \end{figure}
Figure~\ref{fig:d-cliques-example-convergence-speed} illustrates the
performance of D-Cliques on MNIST with $n=100$ nodes. Observe that the Figure~\ref{fig:convergence-speed-dc-vs-fc-2-shards-per-node} illustrates the
convergence speed of D-Cliques with $n=100$ nodes on MNIST (with Clique Averaging) and CIFAR10 (with Clique Averaging and Momentum). Observe that the
convergence speed is convergence speed is
very close very close
to that of a fully-connected topology, and significantly better than with to that of a fully-connected topology, and significantly better than with
a ring or a grid (see Figure~\ref{fig:iid-vs-non-iid-problem}). With a ring or a grid (see Figure~\ref{fig:iid-vs-non-iid-problem}). With
100 nodes, it offers a reduction of $\approx90\%$ in the number of edges 100 nodes, it offers a reduction of $\approx90\%$ in the number of edges
compared to a fully-connected topology. Nonetheless, there is still compared to a fully-connected topology.
significant variance in the accuracy across nodes, which is due to the bias
introduced by inter-clique edges. We address this issue in the next section.
\subsection{D-Cliques Converge Faster than Random Graphs} \subsection{D-Cliques Converge Faster than Random Graphs}
We demonstrate the advantages of D-Cliques over alternative sparse topologies %We demonstrate the advantages of D-Cliques over alternative sparse topologies
that have a similar number of edges. First, we consider topologies in which %that have a similar number of edges. First, we consider topologies in which
the neighbors of each node are selected at random (hence without any clique %the neighbors of each node are selected at random (hence without any clique
structure). %structure).
Specifically, for $n=100$ nodes, we %Specifically, for $n=100$ nodes, we
construct a random topology such that each node has exactly 10 edges, which is %construct a random topology such that each node has exactly 10 edges, which is
similar to the average 9.9 edges of our D-Cliques topology %similar to the average 9.9 edges of our D-Cliques topology
(Figure~\ref{fig:d-cliques-figure}). To better understand the role of %(Figure~\ref{fig:d-cliques-figure}). To better understand the role of
the clique structure beyond merely ensuring class representativity among %the clique structure beyond merely ensuring class representativity among
neighbors, %neighbors,
we also compare to a random topology similar to the one described above except %we also compare to a random topology similar to the one described above except
that edges are %that edges are
chosen such that each node has neighbors of all possible classes. Finally, we %chosen such that each node has neighbors of all possible classes. Finally, we
also implement an analog of Clique Averaging for these random topologies, %also implement an analog of Clique Averaging for these random topologies,
where all nodes de-bias their gradient based on the class distribution of %where all nodes de-bias their gradient based on the class distribution of
their neighbors. In the latter case, since nodes do not form a clique, each %their neighbors. In the latter case, since nodes do not form a clique, each
node obtains a different average gradient. %node obtains a different average gradient.
%
The results for MNIST and CIFAR10 are shown in %The results for MNIST and CIFAR10 are shown in
Figure~\ref{fig:d-cliques-comparison-to-non-clustered-topologies}. For MNIST, %Figure~\ref{fig:d-cliques-comparison-to-non-clustered-topologies}. For MNIST,
a purely random topology has higher variance and lower convergence speed than %a purely random topology has higher variance and lower convergence speed than
D-Cliques (with or without Clique Averaging), while a random topology with %D-Cliques (with or without Clique Averaging), while a random topology with
class representativity performs similarly as D-Cliques without Clique %class representativity performs similarly as D-Cliques without Clique
Averaging. However and perhaps surprisingly, a random topology with unbiased %Averaging. However and perhaps surprisingly, a random topology with unbiased
gradient performs slightly worse than without it. In any case, D-Cliques with %gradient performs slightly worse than without it. In any case, D-Cliques with
Clique Averaging outperforms all random topologies, showing that the clique %Clique Averaging outperforms all random topologies, showing that the clique
structure has a small but noticeable effect on the average accuracy and %structure has a small but noticeable effect on the average accuracy and
significantly reduces the variance across nodes in this setup. %significantly reduces the variance across nodes in this setup.
%
\begin{figure}[t] %\begin{figure}[t]
\centering % \centering
\begin{subfigure}[b]{0.35\textwidth} % \begin{subfigure}[b]{0.35\textwidth}
% To regenerate the figure, from directory results/mnist %% To regenerate the figure, from directory results/mnist
% python ../../../../Software/non-iid-topology-simulator/tools/plot_convergence.py fully-connected-cliques/all/2021-03-10-10:19:44-CET no-init-no-clique-avg/fully-connected-cliques/all/2021-03-12-11:12:49-CET random-10/all/2021-07-23-11:59:56-CEST random-10-diverse/all/2021-03-17-20:28:35-CET --labels 'd-clique (fcc)' 'd-clique (fcc) no clique avg.' '10 random edges' '10 random edges (all classes represented)' --add-min-max --legend 'lower right' --ymin 80 --ymax 92.5 --yaxis test-accuracy --save-figure ../../figures/d-cliques-mnist-linear-comparison-to-non-clustered-topologies.png --font-size 13 --linestyles 'solid' 'dashed' 'dotted' 'dashdot' %% python ../../../../Software/non-iid-topology-simulator/tools/plot_convergence.py fully-connected-cliques/all/2021-03-10-10:19:44-CET no-init-no-clique-avg/fully-connected-cliques/all/2021-03-12-11:12:49-CET random-10/all/2021-07-23-11:59:56-CEST random-10-diverse/all/2021-03-17-20:28:35-CET --labels 'd-clique (fcc)' 'd-clique (fcc) no clique avg.' '10 random edges' '10 random edges (all classes represented)' --add-min-max --legend 'lower right' --ymin 80 --ymax 92.5 --yaxis test-accuracy --save-figure ../../figures/d-cliques-mnist-linear-comparison-to-non-clustered-topologies.png --font-size 13 --linestyles 'solid' 'dashed' 'dotted' 'dashdot'
\centering % \centering
\includegraphics[width=\textwidth]{../figures/d-cliques-mnist-linear-comparison-to-non-clustered-topologies} % \includegraphics[width=\textwidth]{../figures/d-cliques-mnist-linear-comparison-to-non-clustered-topologies}
\caption{MNIST with Linear Model} % \caption{MNIST with Linear Model}
\end{subfigure} % \end{subfigure}
\hfill % \hfill
% To regenerate the figure, from directory results/cifar10 %% To regenerate the figure, from directory results/cifar10
% python ../../../../Software/non-iid-topology-simulator/tools/plot_convergence.py no-init/fully-connected-cliques/all/2021-03-13-18:32:55-CET no-init-no-clique-avg/fully-connected-cliques/all/2021-03-13-18:34:35-CET random-10/all/2021-07-23-14:33:48-CEST random-10-diverse/all/2021-03-17-20:30:41-CET random-10-diverse-unbiased-gradient/all/2021-03-17-20:31:14-CET --labels 'd-clique (fcc) clique avg.' 'd-clique (fcc) no clique avg.' '10 random edges' '10 random edges (all classes repr.)' '10 random (all classes repr.) with unbiased grad.' --add-min-max --legend 'upper left' --yaxis test-accuracy --save-figure ../../figures/d-cliques-cifar10-linear-comparison-to-non-clustered-topologies.png --ymax 119 --font-size 13 --linestyles 'solid' 'dashed' 'dotted' 'dashdot' 'solid' --markers '' '' '' '' 'o' %% python ../../../../Software/non-iid-topology-simulator/tools/plot_convergence.py no-init/fully-connected-cliques/all/2021-03-13-18:32:55-CET no-init-no-clique-avg/fully-connected-cliques/all/2021-03-13-18:34:35-CET random-10/all/2021-07-23-14:33:48-CEST random-10-diverse/all/2021-03-17-20:30:41-CET random-10-diverse-unbiased-gradient/all/2021-03-17-20:31:14-CET --labels 'd-clique (fcc) clique avg.' 'd-clique (fcc) no clique avg.' '10 random edges' '10 random edges (all classes repr.)' '10 random (all classes repr.) with unbiased grad.' --add-min-max --legend 'upper left' --yaxis test-accuracy --save-figure ../../figures/d-cliques-cifar10-linear-comparison-to-non-clustered-topologies.png --ymax 119 --font-size 13 --linestyles 'solid' 'dashed' 'dotted' 'dashdot' 'solid' --markers '' '' '' '' 'o'
\begin{subfigure}[b]{0.35\textwidth} % \begin{subfigure}[b]{0.35\textwidth}
\centering % \centering
\includegraphics[width=\textwidth]{../figures/d-cliques-cifar10-linear-comparison-to-non-clustered-topologies} % \includegraphics[width=\textwidth]{../figures/d-cliques-cifar10-linear-comparison-to-non-clustered-topologies}
\caption{CIFAR10 with LeNet} % \caption{CIFAR10 with LeNet}
\end{subfigure} % \end{subfigure}
\caption{\label{fig:d-cliques-comparison-to-non-clustered-topologies} Comparison to Non-Clustered Topologies} % \caption{\label{fig:d-cliques-comparison-to-non-clustered-topologies} Comparison to Non-Clustered Topologies}
\end{figure} %\end{figure}
%
On the harder CIFAR10 dataset with a deep convolutional network, the %On the harder CIFAR10 dataset with a deep convolutional network, the
differences are much more dramatic: %differences are much more dramatic:
D-Cliques with Clique Averaging and momentum turns out to be critical for fast %D-Cliques with Clique Averaging and momentum turns out to be critical for fast
convergence. %convergence.
Crucially, all random topologies fail to converge to a good solution. This %Crucially, all random topologies fail to converge to a good solution. This
confirms that our clique structure is important to reduce variance %confirms that our clique structure is important to reduce variance
across nodes and improve the convergence. The difference with the previous %across nodes and improve the convergence. The difference with the previous
experiment seems to be due to both the use of a higher capacity model and to %experiment seems to be due to both the use of a higher capacity model and to
the intrinsic characteristics of the datasets. %the intrinsic characteristics of the datasets.
%
While the previous experiments suggest that our clique structure is %While the previous experiments suggest that our clique structure is
instrumental in obtaining good performance, one may wonder whether %instrumental in obtaining good performance, one may wonder whether
intra-clique full connectivity is actually necessary. %intra-clique full connectivity is actually necessary.
Figure~\ref{fig:d-cliques-intra-connectivity} shows the convergence speed of %Figure~\ref{fig:d-cliques-intra-connectivity} shows the convergence speed of
a D-Cliques topology where cliques have been sparsified by randomly %a D-Cliques topology where cliques have been sparsified by randomly
removing 1 or 5 undirected edges per clique (out of 45). Strikingly, both for MNIST and %removing 1 or 5 undirected edges per clique (out of 45). Strikingly, both for MNIST and
CIFAR10, removing just a single edge from the cliques has a %CIFAR10, removing just a single edge from the cliques has a
significant effect on the %significant effect on the
convergence speed. On CIFAR10, it even entirely negates the %convergence speed. On CIFAR10, it even entirely negates the
benefits of D-Cliques. %benefits of D-Cliques.
%
Overall, these results show that achieving fast convergence on non-IID %Overall, these results show that achieving fast convergence on non-IID
data with sparse topologies requires a very careful design, as we have %data with sparse topologies requires a very careful design, as we have
proposed with D-Cliques. %proposed with D-Cliques.
\subsection{Cliques built with Greedy Swap Converge Significantly Faster than Random Cliques} \subsection{Cliques built with Greedy Swap Converge Significantly Faster than Random Cliques}
% From directory 'results-v2':
% MNIST
% python $TOOLS/analyze/filter.py all --dataset:name mnist --topology:name d-cliques/random-cliques d-cliques/greedy-swap --nodes:name 2-shards-uneq-classes --meta:seed 1 --nodes:nb-nodes 100 | python $TOOLS/analyze/diff.py
% python $TOOLS/analyze/diff.py --rundirs all/2021-09-29-22:12:59-CEST-labostrex114 all/2021-09-28-23:18:49-CEST-labostrex119 --pass-through | python $TOOLS/plot/convergence.py --add-min-max --ymin 80 --ymax 92.5 --yaxis test-accuracy --labels 'd-cliques random' 'd-cliques greedy-swap' --save-figure ../mlsys2022style/figures/convergence-speed-mnist-dc-random-vs-dc-gs-2-shards-per-node.png --linestyles 'solid' 'dashed' --font-size 18
% CIFAR10
% python $TOOLS/analyze/filter.py all --dataset:name cifar10 --topology:name d-cliques/random-cliques d-cliques/greedy-swap --nodes:name 2-shards-eq-classes --meta:seed 1 --nodes:nb-nodes 100 | python $TOOLS/analyze/diff.py
% python $TOOLS/analyze/diff.py --rundirs all/2021-10-04-21:18:33-CEST-labostrex117 all/2021-10-03-19:53:21-CEST-labostrex117 --pass-through | python $TOOLS/plot/convergence.py --add-min-max --ymin 0 --ymax 100 --yaxis test-accuracy --labels 'd-cliques random' 'd-cliques greedy-swap' --save-figure ../mlsys2022style/figures/convergence-speed-cifar10-dc-random-vs-dc-gs-2-shards-per-node.png --linestyles 'solid' 'dashed' --font-size 18
\begin{figure}[htbp]
\centering
\begin{subfigure}[b]{0.23\textwidth}
\centering
\includegraphics[width=\textwidth]{figures/convergence-speed-mnist-dc-random-vs-dc-gs-2-shards-per-node}
\caption{\label{fig:convergence-speed-mnist-dc-random-vs-dc-gs-2-shards-per-node} MNIST}
\end{subfigure}
\hfill
\begin{subfigure}[b]{0.23\textwidth}
\centering
\includegraphics[width=\textwidth]{figures/convergence-speed-cifar10-dc-random-vs-dc-gs-2-shards-per-node}
\caption{\label{fig:convergence-speed-cifar10-dc-random-vs-dc-gs-2-shards-per-node} CIFAR10}
\end{subfigure}
\caption{\label{fig:convergence-speed-dc-random-vs-dc-gs-2-shards-per-node} Convergence Speed of D-Cliques constructed Randomly vs Greedy Swap on 100 Nodes (2 shards/node).}
\end{figure}
\subsection{Clique Averaging and Momentum are Necessary} \subsection{Clique Averaging and Momentum are Necessary}
% To regenerate figure, from results/mnist: % From directory 'results-v2':
% python ../../../../Software/non-iid-topology-simulator/tools/plot_convergence.py fully-connected/all/2021-03-10-09:25:19-CET no-init-no-clique-avg/fully-connected-cliques/all/2021-03-12-11:12:49-CET no-init/fully-connected-cliques/all/2021-03-12-11:12:01-CET --add-min-max --yaxis test-accuracy --labels '100 nodes non-IID fully-connected' '100 nodes non-IID d-cliques w/o clique avg.' '100 nodes d-cliques non-IID w/ clique avg.' --legend 'lower right' --ymin 89 --ymax 92.5 --font-size 13 --save-figure ../../figures/d-clique-mnist-clique-avg.png --linestyles 'solid' 'dashed' 'dotted' % MNIST
\begin{figure}[t] % python $TOOLS/analyze/filter.py all --dataset:name mnist --topology:name d-cliques/greedy-swap --nodes:name 2-shards-uneq-classes --meta:seed 1 --nodes:nb-nodes 100 | python $TOOLS/analyze/diff.py
% python $TOOLS/analyze/diff.py --rundirs all/2021-09-29-03:53:42-CEST-labostrex119 all/2021-09-28-23:18:49-CEST-labostrex119 --pass-through | python $TOOLS/plot/convergence.py --add-min-max --ymin 89 --ymax 92.5 --yaxis test-accuracy --labels 'd-cliques w/o c-avg.' 'd-cliques w/ c-avg.' --save-figure ../mlsys2022style/figures/convergence-speed-mnist-dc-no-c-avg-vs-c-avg-2-shards-per-node.png --linestyles 'solid' 'dashed' --font-size 18
\begin{figure}[htbp]
\centering \centering
\includegraphics[width=0.35\textwidth]{../figures/d-clique-mnist-clique-avg} \includegraphics[width=0.35\textwidth]{figures/convergence-speed-mnist-dc-no-c-avg-vs-c-avg-2-shards-per-node}
\caption{\label{fig:d-clique-mnist-clique-avg} Effect of Clique Averaging on MNIST. Y-axis starts at 89.} \caption{\label{fig:d-clique-mnist-clique-avg} Effect of Clique Averaging on MNIST. Y axis starts at 89.}
\end{figure} \end{figure}
As illustrated in Figure~\ref{fig:d-clique-mnist-clique-avg}, Clique Averaging As illustrated in Figure~\ref{fig:d-clique-mnist-clique-avg}, Clique Averaging
...@@ -206,72 +232,65 @@ fully-connected topology. Note that Clique Averaging induces a small ...@@ -206,72 +232,65 @@ fully-connected topology. Note that Clique Averaging induces a small
additional cost, as gradients additional cost, as gradients
and models need to be sent in two separate rounds of messages. Nonetheless, compared to fully connecting all nodes, the total number of messages is reduced by $\approx 80\%$. and models need to be sent in two separate rounds of messages. Nonetheless, compared to fully connecting all nodes, the total number of messages is reduced by $\approx 80\%$.
\begin{figure}[t]
\centering % CIFAR10
% To regenerate figure, from results/cifar10 % python $TOOLS/analyze/filter.py all --dataset:name cifar10 --topology:name d-cliques/greedy-swap --nodes:name 2-shards-eq-classes --meta:seed 1 --nodes:nb-nodes 100 | python $TOOLS/analyze/diff.py
% python ../../../../Software/non-iid-topology-simulator/tools/plot_convergence.py 1-node-iid/all/2021-03-10-13:52:58-CET no-init-no-clique-avg/fully-connected-cliques/all/2021-03-13-18:34:35-CET no-init-no-clique-avg-no-momentum/fully-connected-cliques/all/2021-03-26-13:47:35-CET/ --legend 'upper right' --add-min-max --labels '1-node IID w/ momentum' '100 nodes non-IID d-cliques w/ momentum' '100 nodes non-IID d-cliques w/o momentum' --font-size 14 --yaxis test-accuracy --save-figure ../../figures/d-cliques-cifar10-momentum-non-iid-effect.png --ymax 100 --linestyles 'solid' 'dashed' 'dotted' % w/o Clique Averaging
\begin{subfigure}[b]{0.35\textwidth} % python $TOOLS/analyze/diff.py --rundirs all/2021-10-03-23:37:42-CEST-labostrex117 all/2021-10-04-03:13:46-CEST-labostrex117 --pass-through | python $TOOLS/plot/convergence.py --add-min-max --ymin 0 --ymax 100 --yaxis test-accuracy --labels 'd-cliques w/o momentum' 'd-cliques w/ momentum' --save-figure ../mlsys2022style/figures/convergence-speed-cifar10-wo-c-avg-no-mom-vs-mom-2-shards-per-node.png --linestyles 'solid' 'dashed' --font-size 18
% w/ Clique Averaging
% python $TOOLS/analyze/diff.py --rundirs all/2021-10-03-16:10:34-CEST-labostrex117 all/2021-10-03-19:53:21-CEST-labostrex117 --pass-through | python $TOOLS/plot/convergence.py --add-min-max --ymin 0 --ymax 100 --yaxis test-accuracy --labels 'd-cliques w/o momentum' 'd-cliques w/ momentum' --save-figure ../mlsys2022style/figures/convergence-speed-cifar10-w-c-avg-no-mom-vs-mom-2-shards-per-node.png --linestyles 'solid' 'dashed' --font-size 18
\begin{figure}[htbp]
\centering
\begin{subfigure}[b]{0.23\textwidth}
\centering \centering
\includegraphics[width=\textwidth]{../figures/d-cliques-cifar10-momentum-non-iid-effect} \includegraphics[width=\textwidth]{figures/convergence-speed-cifar10-wo-c-avg-no-mom-vs-mom-2-shards-per-node}
\caption{\label{fig:d-cliques-cifar10-momentum-non-iid-effect} Without Clique Averaging } \caption{\label{fig:convergence-speed-cifar10-wo-c-avg-no-mom-vs-mom-2-shards-per-node} Without Clique Averaging }
\end{subfigure} \end{subfigure}
\hfill \hfill
% To regenerate figure, from results/cifar10 \begin{subfigure}[b]{0.23\textwidth}
% python ../../../../Software/non-iid-topology-simulator/tools/plot_convergence.py 1-node-iid/all/2021-03-10-13:52:58-CET no-init/fully-connected-cliques/all/2021-03-13-18:32:55-CET --legend 'upper right' --add-min-max --labels '1-node IID w/ momentum' '100 nodes non-IID d-clique w/ momentum' --font-size 14 --yaxis test-accuracy --save-figure ../../figures/d-cliques-cifar10-momentum-non-iid-clique-avg-effect.png --ymax 100 --linestyles 'solid' 'dashed' 'dotted'
\begin{subfigure}[b]{0.35\textwidth}
\centering \centering
\includegraphics[width=\textwidth]{../figures/d-cliques-cifar10-momentum-non-iid-clique-avg-effect} \includegraphics[width=\textwidth]{figures/convergence-speed-cifar10-w-c-avg-no-mom-vs-mom-2-shards-per-node}
\caption{\label{fig:d-cliques-cifar10-momentum-non-iid-clique-avg-effect} With Clique Averaging} \caption{\label{fig:convergence-speed-cifar10-w-c-avg-no-mom-vs-mom-2-shards-per-node} With Clique Averaging}
\end{subfigure} \end{subfigure}
\caption{\label{fig:cifar10-momentum} Non-IID Effect of Momentum on CIFAR10 with LeNet} \caption{\label{fig:cifar10-c-avg-momentum} Effect of Clique Averaging and Momentum on CIFAR10 with LeNet.}
\end{figure} \end{figure}
As shown in As shown in
Figure~\ref{fig:d-cliques-cifar10-momentum-non-iid-clique-avg-effect}, Figure~\ref{fig:cifar10-c-avg-momentum},
the use of Clique Averaging restores the benefits of momentum and closes the gap the use of Clique Averaging restores the benefits of momentum and closes the gap
with the centralized setting. with the centralized setting.
\subsection{Full Intraclique Connectivity is Necessary} \subsection{D-Cliques Tolerate Intra-Connectivity Failures}
\begin{figure*}[t] % From directory 'results-v2':
% MNIST
% python $TOOLS/analyze/filter.py all --dataset:name mnist --topology:name d-cliques/greedy-swap --nodes:name 2-shards-uneq-classes --meta:seed 1 --nodes:nb-nodes 100 | python $TOOLS/analyze/diff.py
% w/o Clique Gradient
% python $TOOLS/analyze/diff.py --rundirs all/2021-09-29-03:53:42-CEST-labostrex119 all/2021-10-01-21:44:14-CEST-labostrex113 all/2021-10-02-06:53:40-CEST-labostrex113 --pass-through | python $TOOLS/plot/convergence.py --add-min-max --ymin 89 --ymax 92.5 --yaxis test-accuracy --labels 'full intra-connectivity' '-1 edge/clique' '-5 edges/clique' --save-figure ../mlsys2022style/figures/d-cliques-wo-clique-avg-impact-of-edge-removal.png --linestyles 'solid' 'dashed' 'dotted' --font-size 18
% w/ Clique Gradient
% python $TOOLS/analyze/diff.py --rundirs all/2021-09-28-23:18:49-CEST-labostrex119 all/2021-10-01-17:08:42-CEST-labostrex113 all/2021-10-02-02:17:43-CEST-labostrex113 --pass-through | python $TOOLS/plot/convergence.py --add-min-max --ymin 89 --ymax 92.5 --yaxis test-accuracy --labels 'full intra-connectivity' '-1 edge/clique' '-5 edges/clique' --save-figure ../mlsys2022style/figures/d-cliques-w-clique-avg-impact-of-edge-removal.png --linestyles 'solid' 'dashed' 'dotted' --font-size 18
\begin{figure}[htbp]
\centering \centering
\begin{subfigure}[htbp]{0.4\textwidth} \begin{subfigure}[htbp]{0.23\textwidth}
\centering \centering
% To regenerate the figure, from directory results/mnist \includegraphics[width=\textwidth]{figures/d-cliques-wo-clique-avg-impact-of-edge-removal}
% python ../../../../Software/non-iid-topology-simulator/tools/plot_convergence.py no-init/fully-connected-cliques/all/2021-03-12-11:12:01-CET rm-1-edge/all/2021-03-18-17:28:27-CET rm-1-edge-unbiased-grad/all/2021-03-18-17:28:47-CET --add-min-max --ymin 85 --ymax 92.5 --legend 'lower right' --yaxis test-accuracy --labels 'fcc, clique grad.' 'fcc -1 edge/clique, no clique grad.' 'fcc -1 edge/clique, clique grad.' --save-figure ../../figures/d-cliques-mnist-clique-clustering-fcc-minus-1-edge.png --font-size 13 --linestyle 'solid' 'dashed' 'dotted' \caption{\label{fig:d-cliques-wo-clique-avg-impact-of-edge-removal} Without Clique Averaging }
\includegraphics[width=\textwidth]{../figures/d-cliques-mnist-clique-clustering-fcc-minus-1-edge}
\caption{\label{fig:d-cliques-mnist-clique-clustering-minus-1-edge} MNIST (-1 edge/clique)}
\end{subfigure} \end{subfigure}
\hfill \hfill
\begin{subfigure}[htbp]{0.4\textwidth} \begin{subfigure}[htbp]{0.23\textwidth}
\centering \centering
% To regenerate the figure, from directory results/cifar10 \includegraphics[width=\textwidth]{figures/d-cliques-w-clique-avg-impact-of-edge-removal}
% python ../../../../Software/non-iid-topology-simulator/tools/plot_convergence.py no-init/fully-connected-cliques/all/2021-03-13-18:32:55-CET rm-1-edge/all/2021-03-18-17:29:58-CET rm-1-edge-unbiased-grad/all/2021-03-18-17:30:17-CET --add-min-max --ymax 80 --legend 'upper left' --yaxis test-accuracy --labels 'fcc, clique grad.' 'fcc -1 edge/clique, no clique grad.' 'fcc -1 edge/clique, clique grad.' --save-figure ../../figures/d-cliques-cifar10-clique-clustering-fcc-minus-1-edge.png --font-size 13 --linestyle 'solid' 'dashed' 'dotted' \caption{\label{fig:d-cliques-w-clique-avg-impact-of-edge-removal} With Clique Averaging}
\includegraphics[width=\textwidth]{../figures/d-cliques-cifar10-clique-clustering-fcc-minus-1-edge}
\caption{\label{fig:d-cliques-cifar10-clique-clustering-minus-1-edge} CIFAR10 (-1 edge/clique)}
\end{subfigure} \end{subfigure}
\caption{\label{fig:d-cliques-intra-connectivity} MNIST: Impact of Intra-clique Connectivity Failures. Y axis starts at 89.}
\end{figure}
%\begin{subfigure}[htbp]{0.35\textwidth} % CIFAR10
% \centering % python $TOOLS/analyze/filter.py all --dataset:name cifar10 --topology:name d-cliques/greedy-swap --nodes:name 2-shards-eq-classes --meta:seed 1 --nodes:nb-nodes 100 | python $TOOLS/analyze/diff.py
%% To regenerate the figure, from directory results/mnist
%% python ../../../../Software/non-iid-topology-simulator/tools/plot_convergence.py no-init/fully-connected-cliques/all/2021-03-12-11:12:01-CET rm-5-edges/all/2021-03-18-17:29:10-CET rm-5-edges-unbiased-grad/all/2021-03-18-17:29:36-CET --add-min-max --ymin 85 --ymax 92.5 --legend 'lower right' --yaxis test-accuracy --labels 'fcc, clique grad.' 'fcc -5 edges/clique, no clique grad.' 'fcc -5 edges/clique, clique grad.' --save-figure ../../figures/d-cliques-mnist-clique-clustering-fcc-minus-5-edges.png --font-size 13 --linestyle 'solid' 'dashed' 'dotted'
% \includegraphics[width=\textwidth]{../figures/d-cliques-mnist-clique-clustering-fcc-minus-5-edges}
%\caption{\label{fig:d-cliques-mnist-clique-clustering-minus-5-edges} MNIST (-5 edges/clique)}
%\end{subfigure}
%\hfill
%\begin{subfigure}[htbp]{0.35\textwidth}
% \centering
%% To regenerate the figure, from directory results/cifar10
%% python ../../../../Software/non-iid-topology-simulator/tools/plot_convergence.py no-init/fully-connected-cliques/all/2021-03-13-18:32:55-CET rm-5-edges/all/2021-03-18-17:30:38-CET rm-5-edges-unbiased-grad/all/2021-03-18-17:31:04-CET --add-min-max --ymax 80 --legend 'upper left' --yaxis test-accuracy --labels 'fcc, clique grad.' 'fcc -5 edges/clique, no clique grad.' 'fcc -5 edges/clique, clique grad.' --save-figure ../../figures/d-cliques-cifar10-clique-clustering-fcc-minus-5-edges.png --font-size 13 --linestyle 'solid' 'dashed' 'dotted'
% \includegraphics[width=\textwidth]{../figures/d-cliques-cifar10-clique-clustering-fcc-minus-5-edges}
%\caption{\label{fig:d-cliques-cifar10-clique-clustering-minus-5-edges} CIFAR10 (-5 edges/clique)}
%\end{subfigure}
\caption{\label{fig:d-cliques-intra-connectivity} Importance of Intra-Clique Full-Connectivity}
\end{figure*}
\subsection{D-Cliques appear to Scale with Sparser Inter-Clique Topologies} \subsection{D-Cliques Scale with Sparser Inter-Clique Topologies}
In this last series of experiments, we evaluate the effect of choosing sparser In this last series of experiments, we evaluate the effect of choosing sparser
inter-clique topologies on the convergence speed for a larger network of 1000 inter-clique topologies on the convergence speed for a larger network of 1000
......
mlsys2022style/figures/convergence-speed-cifar10-dc-fc-vs-fc-1-class-per-node.png 0 → 100644
+ 0
− 0
View file @ 1ca8944e
mlsys2022style/figures/convergence-speed-cifar10-dc-fc-vs-fc-1-class-per-node.png

45.3 KiB

mlsys2022style/figures/convergence-speed-cifar10-dc-fc-vs-fc-2-shards-per-node.png 0 → 100644
+ 0
− 0
View file @ 1ca8944e
mlsys2022style/figures/convergence-speed-cifar10-dc-fc-vs-fc-2-shards-per-node.png

43.9 KiB

mlsys2022style/figures/convergence-speed-mnist-dc-fc-vs-fc-1-class-per-node.png 0 → 100644
+ 0
− 0
View file @ 1ca8944e
mlsys2022style/figures/convergence-speed-mnist-dc-fc-vs-fc-1-class-per-node.png

42.7 KiB

mlsys2022style/figures/convergence-speed-mnist-dc-fc-vs-fc-2-shards-per-node.png 0 → 100644
+ 0
− 0
View file @ 1ca8944e
mlsys2022style/figures/convergence-speed-mnist-dc-fc-vs-fc-2-shards-per-node.png

44.6 KiB

mlsys2022style/figures/convergence-speed-mnist-dc-random-vs-dc-gs-2-shards-per-node.png 0 → 100644
+ 0
− 0
View file @ 1ca8944e
mlsys2022style/figures/convergence-speed-mnist-dc-random-vs-dc-gs-2-shards-per-node.png

56.5 KiB

mlsys2022style/figures/convergence-speed-mnist-dc-rm-1-edge-vs-full.png 0 → 100644
+ 0
− 0
View file @ 1ca8944e
mlsys2022style/figures/convergence-speed-mnist-dc-rm-1-edge-vs-full.png

60.7 KiB

mlsys2022style/figures/d-cliques-ideal-w-clique-avg-impact-of-edge-removal.png 0 → 100644
+ 0
− 0
View file @ 1ca8944e
mlsys2022style/figures/d-cliques-ideal-w-clique-avg-impact-of-edge-removal.png

78.3 KiB

mlsys2022style/figures/d-cliques-ideal-wo-clique-avg-impact-of-edge-removal.png 0 → 100644
+ 0
− 0
View file @ 1ca8944e
mlsys2022style/figures/d-cliques-ideal-wo-clique-avg-impact-of-edge-removal.png

90.5 KiB

mlsys2022style/figures/d-cliques-w-clique-avg-impact-of-edge-removal.png 0 → 100644
+ 0
− 0
View file @ 1ca8944e
mlsys2022style/figures/d-cliques-w-clique-avg-impact-of-edge-removal.png

72.4 KiB

mlsys2022style/figures/d-cliques-wo-clique-avg-impact-of-edge-removal.png 0 → 100644
+ 0
− 0
View file @ 1ca8944e
mlsys2022style/figures/d-cliques-wo-clique-avg-impact-of-edge-removal.png

78.3 KiB

0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment