Spectral Clustering vs. K-means (Pavlovic)

[dlee138]

What was the strategic reason as to why spectral clustering was used in the Pavlovic paper? Can someone give me a brief overview of how spectral clustering works and how it differs from some of the other methods we discussed in class such as k-means?

[ajulian3]

K Means deals with compactness of spatial data while spectral clustering deals with connectivity. Spectral Clustering is important when determining neural network connectivity. I would think K means is important when determining more of a geographical localization understanding. Does this make more sense?

[dlee138]

Yes, so K-means focuses more on "closeness" while spectral deals with connectivity.

[michaelseung]

I think spectral clustering is also more useful when dealing with higher dimensions with the use of eigenvalues to cluster in fewer dimensions.