150 lines
2.5 KiB
Plaintext
150 lines
2.5 KiB
Plaintext
|
<subsection Graph Autoencoder>
|
||
|
|
<frame>
|
||
|
|
|
||
|
|
<list>
|
||
|
|
<e>So here ParticleNet + QCDorWhat</e>
|
||
|
|
<e>=Graph Autoencoder</e>
|
||
|
|
<e>To do this, require:</e>
|
||
|
|
<l2st>
|
||
|
|
<e>An update step</e>
|
||
|
|
<l3st><e>already described</e></l3st>
|
||
|
|
<ignore>
|
||
|
|
<e>Something to make a list of 4 vectors into a graph</e>
|
||
|
|
<l3st>
|
||
|
|
<e>use a topK algorithm</e>
|
||
|
|
<e>connect a 4 vector with its K nearest neigbours</e>
|
||
|
|
</l3st>
|
||
|
|
</ignore>
|
||
|
|
<e>something to reduce the number of nodes</e>
|
||
|
|
<e>something to increase the number of nodes afterwards again</e>
|
||
|
|
</l2st>
|
||
|
|
</list>
|
||
|
|
|
||
|
|
</frame>
|
||
|
|
<frame title="Going from a big graph to a small graph">
|
||
|
|
<list>
|
||
|
|
<e>similar to a Pooling operation for a convolutional network</e>
|
||
|
|
<e>Seems simple enough but if you look at the literature</e>
|
||
|
|
<l2st>
|
||
|
|
<e>slow...and the benefits...are less clear (arXiv:1907.09000)</e>
|
||
|
|
<e>advance...has lagged behind (arXiv:1907.00481)</e>
|
||
|
|
<e>one cannot simply pool ... (arXiv:1806.08804)</e>
|
||
|
|
</l2st>
|
||
|
|
</list>
|
||
|
|
|
||
|
|
</frame>
|
||
|
|
<frame>
|
||
|
|
<split>
|
||
|
|
<que>
|
||
|
|
<i f="xdenseencode.png"/>
|
||
|
|
</que>
|
||
|
|
<que>
|
||
|
|
<i f="xgraphencode.png"/>
|
||
|
|
</que>
|
||
|
|
</split>
|
||
|
|
<ignore>
|
||
|
|
<i f="xdenseencode.png" f2="xgraphencode.png"></i>
|
||
|
|
</ignore>
|
||
|
|
</frame>
|
||
|
|
<frame title="from a big graph to a small graph">
|
||
|
|
|
||
|
|
<split>
|
||
|
|
<que>
|
||
|
|
<list>
|
||
|
|
<e>project the graph on a learnable axis</e>
|
||
|
|
<e>combine neigbour nodes on this axis</e>
|
||
|
|
<e>relearn the graph or use a graph combination rule</e>
|
||
|
|
</list>
|
||
|
|
</que>
|
||
|
|
<que>
|
||
|
|
<i f="abiaa" wmode=True></i>
|
||
|
|
<ignore>maybe instead of multiple images below each other doo this on multiple pages</ignore>
|
||
|
|
</que>
|
||
|
|
</split>
|
||
|
|
|
||
|
|
</frame>
|
||
|
|
<frame title="from a small graph to a big graph">
|
||
|
|
|
||
|
|
<split>
|
||
|
|
<que>
|
||
|
|
<list>
|
||
|
|
<e>let each node grow into a learnable graph</e>
|
||
|
|
<e>combine the new graphs with the existing one</e>
|
||
|
|
</list>
|
||
|
|
</que>
|
||
|
|
<que>
|
||
|
|
<i f="abibb" wmode=True></i>
|
||
|
|
</que>
|
||
|
|
</split>
|
||
|
|
|
||
|
|
</frame>
|
||
|
|
|
||
|
|
<ignore>
|
||
|
|
|
||
|
|
<frame><i f="../../mmeet2/imgs/00.pdf"/></frame>
|
||
|
|
<frame><i f="../../mmeet2/imgs/01.pdf"/></frame>
|
||
|
|
<frame><i f="../../mmeet2/imgs/02.pdf"/></frame>
|
||
|
|
<frame><i f="../../mmeet2/imgs/03.pdf"/></frame>
|
||
|
|
<frame><i f="../../mmeet2/imgs/04.pdf"/></frame>
|
||
|
|
<frame><i f="../../mmeet2/imgs/05.pdf"/></frame>
|
||
|
|
<frame><i f="../../mmeet2/imgs/06.pdf"/></frame>
|
||
|
|
<frame><i f="../../mmeet2/imgs/07.pdf"/></frame>
|
||
|
|
<frame><i f="../../mmeet2/imgs/08.pdf"/></frame>
|
||
|
|
<frame><i f="../../mmeet2/imgs/06.pdf"/></frame>
|
||
|
|
<frame><i f="../../mmeet2/imgs/01.pdf"/></frame>
|
||
|
|
|
||
|
|
</ignore>
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
<ignore>
|
||
|
|
|
||
|
|
<split>
|
||
|
|
<que>
|
||
|
|
<list>
|
||
|
|
<e></e>
|
||
|
|
<e></e>
|
||
|
|
<e></e>
|
||
|
|
</list>
|
||
|
|
</que>
|
||
|
|
<que>
|
||
|
|
|
||
|
|
</que>
|
||
|
|
</split>
|
||
|
|
|
||
|
|
<frame>
|
||
|
|
|
||
|
|
<split>
|
||
|
|
<que>
|
||
|
|
<list>
|
||
|
|
<e></e>
|
||
|
|
<e></e>
|
||
|
|
<e></e>
|
||
|
|
</list>
|
||
|
|
</que>
|
||
|
|
<que>
|
||
|
|
<i f="none"></i>
|
||
|
|
</que>
|
||
|
|
</split>
|
||
|
|
|
||
|
|
</frame>
|
||
|
|
|
||
|
|
<frame>
|
||
|
|
|
||
|
|
<list>
|
||
|
|
<e></e>
|
||
|
|
<e></e>
|
||
|
|
<e></e>
|
||
|
|
</list>
|
||
|
|
|
||
|
|
</frame>
|
||
|
|
|
||
|
|
|
||
|
|
</ignore>
|
||
|
|
|
||
|
|
|
||
|
|
|