<html dir="ltr">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
<style type="text/css" id="owaParaStyle"></style>
</head>
<body fpstyle="1" ocsi="0">
<div style="direction: ltr;font-family: Tahoma;color: #000000;font-size: 10pt;">Dear all,
<div><br>
</div>
<div>I wanted to make you aware of an excellent resource that is currently being compiled by Monica Bobra and James Mason: the HelioML e-book (<a href="https://helioml.github.io/HelioML/chapters/title" target="_blank">https://helioml.github.io/HelioML/chapters/title</a>). </div>
<div><span style="font-size: 10pt;"><br>
</span></div>
<div><span style="font-size: 10pt;">We have recently contributed a chapter based on
</span><span style="font-size: 10pt;">DeepEM</span><span style="font-size: 10pt;">: a (supervised) deep learning approach to DEM inversion. DeepEM is currently under development on Github (https://github.com/PaulJWright/DeepEM), and </span><span style="font-size: 10pt;">DeepEM
has been designed to both accelerate the computation of DEM solutions, and reduce erroneous solutions provided by
</span><i style="font-size: 10pt;">pre-existing</i><span style="font-size: 10pt;"> inversion methods. The underlying idea is that a neural network should be able to learn a generalized mapping </span><span style="font-size: 10pt;">in order to transform an SDO/AIA
input vector (6x1x1) to the DEM output vector (Nx1x1, where N is the number of temperature bins of the DEM solution) also retain the fidelity to the original inversion method.</span></div>
<div><span style="font-size: 10pt;"><br>
</span></div>
<div><span style="font-size: 10pt;">In our DeepEM chapter we provide a simple model in order to demonstrate how easy it is to apply deep learning techniques to your own research. This very simple implementation is trained on one</span><span style="font-size: 13.3333px;"> set
of 512x512 AIA images, and corresponding Basis Pursuit DEM solutions. This implementation returns</span><span style="font-size: 10pt;"> >10 Million DEM/s with similar fidelity to Basis Pursuit. </span><span style="font-size: 10pt;">A gif of how this implementation
compares to Basis Pursuit is shown here: </span><a href="https://twitter.com/PaulJWright/status/1068115587614863360," target="_blank" style="font-size: 10pt;">https://twitter.com/PaulJWright/status/1068115587614863360</a>, and<span style="font-size: 10pt;"> </span><span style="font-size: 10pt;">I
hope that this chapter, and e-book in general, will encourage those who are interested to applying deep learning to their research.</span></div>
<div><br>
</div>
<div>
<div>Cheers,</div>
<div>Paul<br>
<div style="font-family:Tahoma; font-size:13px">
<div style="font-family:Tahoma; font-size:13px">
<div style="font-size:13px; font-family:Tahoma">---<br>
<div style="orphans:2; widows:2">Paul J. Wright<br>
</div>
<div style="orphans:2; widows:2">Research Assistant & Ph.D. Student<br>
<br>
</div>
<div style="orphans:2; widows:2">Astronomy & Astrophysics Group</div>
<div style="orphans:2; widows:2">SUPA School of Physics & Astronomy</div>
<div style="orphans:2; widows:2">Kelvin Building</div>
<div style="orphans:2; widows:2">University of Glasgow</div>
<div style="orphans:2; widows:2">Glasgow, UK <br>
G12 8QQ</div>
<div style="orphans:2; widows:2"><br>
Web: <a href="http://www.pauljwright.co.uk">http://www.pauljwright.co.uk</a></div>
<div style="orphans:2; widows:2">Email: <a href="mailto:paul.wright@glasgow.ac.uk">
paul.wright@glasgow.ac.uk</a><br>
Email (<span class="spell">permanent</span>): <a href="mailto:paul@pauljwright.co.uk">
paul@pauljwright.co.uk<br>
</a></div>
</div>
</div>
</div>
</div>
</div>
</div>
</body>
</html>