Zoom: https://uio.zoom.us/j/8621305603
Agenda
From the "share your research" part we will have a short talk by Thore:
"Clustering of spectral profiles: can you trust k-means?"
K-means is a very popular technique for clustering spectral profiles. It
works by grouping together profiles with similar features around a mean
profile, which ideally will be representative of the cluster members.
For example, these mean profiles are used to speed up inversions, or for
detecting the presence and location of particular phenomena in
observations. However, k-means is not inherently well suited for dealing
with spectral profiles, and there are several pitfalls one should be
aware of. Since several of us currently are, have been or will be using
k-means, I would like to talk to you about these potential gotcha's, as
well as introduce an alternative technique called k-shape which, at
least in some cases, can sidestep the problems of k-means.
In addition, we will also hear from Reetika about the recent Hinode-15 /
IRIS-12 meeting in Prague.