BEGIN:VCALENDAR
VERSION:2.0
CALSCALE:GREGORIAN
PRODID:UW-Madison-Physics-Events
BEGIN:VEVENT
SEQUENCE:1
UID:UW-Physics-Event-6181
DTSTART:20210127T170000Z
DTEND:20210127T181500Z
DTSTAMP:20260415T023559Z
LAST-MODIFIED:20210118T165258Z
LOCATION:Online Seminar: Please sign up for our mailing list at www.ph
 ysicsmeetsml.org for zoom link
SUMMARY:The Importance of Being Interpretable\, Physics ∩ ML Seminar
 \, Michelle Ntampaka\, Space Telescope Science Institute
DESCRIPTION:Cosmology is entering an era of data-driven science. Moder
 n machine learning techniques are being combined with large astronomic
 al surveys to enable powerful new research methods. This shift in our 
 scientific approach requires us to ask an important question: Can we t
 rust the black box?
\n
\nI will present a deep machine learning approa
 ch to constraining cosmological parameters with multi-wavelength obser
 vations of galaxy clusters. The ML approach has two components: an aut
 oencoder that builds a compressed representation of each galaxy cluste
 r and a flexible CNN to estimate the cosmological model from a cluster
  sample. From mock observations\, the ML method estimates the amplitud
 e of matter fluctuations\, sigma8\, at approximately the expected theo
 retical limit. More importantly\, the deep ML approach can be understo
 od and interpreted. I will lay out three schemes for interpreting the 
 ML technique: a leave-one-out method for assessing cluster importance\
 , an average saliency for evaluating feature importance\, and correlat
 ions in the terse layer for understanding whether an ML technique can 
 be safely applied to observational data. I will introduce the term “
 overspecialized" to describe a common pitfall in astronomical applicat
 ions of machine learning in which the ML method learns simulation-spec
 ific details\, and we show how a carefully sculpted architecture can b
 e used to check for this source of systematic error.
URL:https://www.physics.wisc.edu/events/?id=6181
END:VEVENT
END:VCALENDAR