<html>
<head>
<meta http-equiv="content-type" content="text/html; charset=utf-8">
</head>
<body bgcolor="#FFFFFF" text="#000000">
<div
style="font-family:arial,sans-serif;font-size:13.3333339691162px">=================================================================<br>
</div>
<div
style="font-family:arial,sans-serif;font-size:13.3333339691162px"><br>
</div>
<div>Call for Papers: <i>Modern Nonparametrics 3: Automating the
Learning Pipeline</i><i><span
style="color:rgb(0,0,0);font-family:inherit;font-size:13px;line-height:16.4133319854736px;white-space:pre-wrap"></span></i><br>
held in conjunction with Neural Information Processing Systems
(NIPS 2014)</div>
<div><br>
</div>
<div>December 13, 2014, Montreal, Canada</div>
<div><br>
</div>
<div><a
href="https://sites.google.com/site/nips2014modernnonparametric/"
target="_blank">https://sites.google.com/site/nips2014modernnonparametric/</a><br>
</div>
<div><br>
</div>
<div>--------------------------------------------------------------------------------------------------------------<br>
</div>
<div><br>
</div>
<div>Overview:</div>
<div>-------------</div>
<div><br>
</div>
<div>
<div>Nonparametric methods (kernel methods, kNN, classification
trees, etc) are</div>
<div>designed to handle complex pattern recognition problems.
Such complex</div>
<div>problems arise in modern applications such as genomic
experiments, climate</div>
<div>analysis, robotic control, social network analysis, and so
forth. </div>
<div>There is a growing need for statistical procedures that can
be used</div>
<div>“off-the-shelf”, i.e. procedures with as few parameters as
possible, or</div>
<div>better yet, procedures which can “self-tune” to a particular
application</div>
<div>at hand.</div>
<div><br>
</div>
<div>In traditional statistics, much effort has gone into so
called</div>
<div>“adaptive” procedures which can attain optimal risks over
large sets of</div>
<div>models of increasing complexity. Examples are model selection
approaches</div>
<div>based on penalized empirical risk minimization, approaches
based on</div>
<div>stability of estimates (e.g. Lepski’s methods), thresholding
approaches</div>
<div>under sparsity assumptions, and model averaging approaches.
Most of these</div>
<div>approaches rely on having tight bounds on the risk of
learning procedures</div>
<div>(under any parameter setting), hence other approaches
concentrate on tight</div>
<div>estimations of the actual risks, e.g., Stein’s risk
estimators,</div>
<div>bootstrapping methods, data dependent learning bounds.</div>
<div><br>
</div>
<div>In theoretical machine learning, much of the work has focused
on proper</div>
<div>tuning of the actual optimization procedures used to minimize
(penalized)</div>
<div>empirical risks. In particular, great effort has gone into
the automatic</div>
<div>setting of important tuning parameters such as ‘learning
rates’ and ‘step</div>
<div>sizes’.</div>
<div><br>
</div>
<div>Another approach out of machine learning arises in the kernel
literature</div>
<div>under the name of ‘automatic representation learning’. The
aim of the</div>
<div>approach, similar to theoretical work on model selection, is
to</div>
<div>automatically learn an appropriate (kernel) transformation of
the data for</div>
<div>use with kernel methods such as SVMs or Gaussian processes.</div>
<div><br>
</div>
<div>A main aim of this workshop is to cover the various
approaches proposed so</div>
<div>far towards automating the learning pipeline, and the
practicality of these</div>
<div>approaches in light of modern constraints. We are
particularly interested</div>
<div>in understanding whether large datasizes and dimensionality
might</div>
<div>help the automation effort since such datasets in fact
provide more</div>
<div>information on the patterns being learned.</div>
<div><br>
</div>
<div>This workshop is third in a series of NIPS workshops on
modern</div>
<div>nonparametric methods in machine learning, which several of
the present</div>
<div>organizers were involved in running during NIPS 2013 and NIPS
2012 (see</div>
<div>organizer biographies). These previous workshops focused on
the challenges</div>
<div>posed by large data sizes (e.g. time/accuracy tradeoffs) and
large</div>
<div>dimensionality (e.g. dimension reduction strategies). The
main focus of the <br>
present workshop, automating the learning pipeline, builds on
these</div>
<div>previous workshops.</div>
</div>
<div><br>
</div>
<div><br>
</div>
<div>Submission:</div>
<div>----------------</div>
<div><br>
</div>
<div>Papers submitted to the workshop should be up to four pages
long, extended <br>
abstracts in camera-ready format using the NIPS style. They should
be sent <br>
by email to ''<a href="mailto:nonparametric.nips2014@gmail.com"
target="_blank">nonparametric.nips2014@gmail.com</a>''. Accepted
submissions will <br>
be presented as talks or posters.<br>
</div>
<div><br>
</div>
<div>Important Dates:</div>
<div>-----------------------</div>
<div><br>
</div>
<div>submission deadline: Oct 9, 2014 (23:59 UTC)</div>
<div>notification of acceptance: Oct 23, 2014 (23:59 UTC)</div>
<div>workshop: Dec 13, 2014</div>
<div><br>
</div>
<div>Registration: </div>
<div>-----------------</div>
<div><br>
</div>
Participants should refer to the NIPS-2014 website for information
on how to <br>
register for the workshop.<br>
<br>
<pre class="moz-signature" cols="72">--
Zoltan Szabo
Gatsby Computational Neuroscience Unit
University College London
<a class="moz-txt-link-freetext" href="http://www.gatsby.ucl.ac.uk/%7Eszabo/">http://www.gatsby.ucl.ac.uk/~szabo/</a>
</pre>
<br>
</body>
</html>