<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=Windows-1252">
<style type="text/css" style="display:none;"> P {margin-top:0;margin-bottom:0;} </style>
</head>
<body dir="ltr">
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0); background-color: rgb(255, 255, 255);">
<span style="margin: 0px; font-size: 15px; color: rgb(0, 0, 0); background-color: white;">The AI Lunch Series will be hosting Dr. João Caldeira (Google) this Wednesday for a presentation on “<i style="font-family:Calibri, sans-serif;font-size:11pt;text-align:center"><span style="margin:0px;font-size:12pt;font-family:"Calibri Light", sans-serif">Deeply
Uncertain: Comparing Methods of Uncertainty<span> </span></span></i><i style="text-align:center"><span style="margin:0px;font-size:12pt;line-height:17.12px;font-family:"Calibri Light", sans-serif"><i><span style="margin:0px">Quantification in Deep Learning
Algorithms</span></i></span></i><span style="margin:0px;font-family:Calibri, sans-serif;text-align:center">”. Additional information can be found at: </span><a href="https://www.jlab.org/AI/lunch_series/2021" target="_blank" rel="noopener noreferrer" data-auth="NotApplicable" style="margin:0px;font-family:Calibri, sans-serif;text-align:center">https://www.jlab.org/AI/lunch_series/2021</a></span>
<div style="margin: 0px; font-size: 15px; color: rgb(0, 0, 0); background-color: white;">
<br>
</div>
<div style="margin: 0px; font-size: 15px; color: rgb(0, 0, 0); background-color: white;">
12-1PM, Wednesday, February 17, 2021</div>
<div style="margin: 0px; font-size: 15px; color: rgb(0, 0, 0); background-color: white;">
Join with Bluejeans:<span style="margin:0px"> <a href="https://bluejeans.com/950395297" target="_blank" rel="noopener noreferrer" data-auth="NotApplicable" style="margin:0px">https://bluejeans.com/950395297</a></span></div>
<div style="margin: 0px; font-size: 15px; color: rgb(0, 0, 0); background-color: white;">
<br>
</div>
<div style="margin: 0px; font-size: 15px; color: rgb(0, 0, 0); background-color: white;">
Abstract: In recent years, many methods for quantifying uncertainty in the output of deep learning algorithms have been developed. In this talk we will introduce three of the most prominent methods - Bayesian Neural Networks (BNN), Concrete Dropout (CD), and
Deep Ensembles (DE) - and compare them to the standard analytic error propagation in the context of a single pendulum. We will also relate the output of these methods to terms more familiar in the context of the physical sciences. Our results highlight some
pitfalls that may occur when using these UQ methods. For example, when the variation of noise in the training set is small, all methods predicted the same relative uncertainty independently of the inputs. This issue is particularly hard to avoid in BNN. On
the other hand, when the test set contains samples far from the training distribution, we found that no methods sufficiently increased the uncertainties associated to their predictions. This problem was particularly clear for CD. In light of these results,
we make some recommendations for usage and interpretation of UQ methods.<br>
</div>
<div style="margin: 0px; font-size: 15px; color: rgb(0, 0, 0); background-color: white;">
<br>
</div>
<span style="margin: 0px; font-size: 15px; color: rgb(0, 0, 0); background-color: white;">If you have questions, contact tennant@jlab.org </span><br>
</div>
</body>
</html>