<html>
<head>
<meta http-equiv="content-type" content="text/html; charset=UTF-8">
</head>
<body text="#000000" bgcolor="#99FF99">
<font size="-1"><font face="Helvetica, Arial, sans-serif">This is an
update of a post a few months back, where we introduced a
generalization of<br>
DROPOUT for DLs called the Stochastic Delta Rule. We have
updated now all properly tested benchmarks which still show a
significant improvement in all cases for SDR in both accuracy
and speed over DROPOUT. We have added more benchmarks
including for ImageNet and updated the paper which is attached
to this post.<br>
<br>
The Github is also completely updated with updated PyTorch code
and results and relevant usage comments here:<br>
</font></font><br>
<font size="-1"><font face="Helvetica, Arial, sans-serif"> <a
href="https://github.com/noahfl/sdr-densenet-pytorch">https://github.com/noahfl/sdr-densenet-pytorch</a><br>
</font></font>
<pre class="moz-signature" cols="72">--
Stephen José Hanson
Professor
Director RUBIC (University-Wide)
Department of Psychology (NK)
Cognitive Science Center (NB)</pre>
</body>
</html>