We lift up each Schreier eigenspace
To over-span the data
These vectors form a tight group frame
To analyze the data
What’s a tight group frame
Oh, Why a tight group frame?
The atoms are interpretable
When dotted with the data
Each atom of the tight group frame
Is a portion of the data
It picks out a pattern of voting types
To find trends in the data
We can lift each node from a Schreier coset graph (at left) to the full permutahedron based on different permutations of candidates in a tableau. At right, there are three unique liftings, which give us 3 vectors to span a 2 dimensional space. Each of the Schreier graphs corresponds to a clustering of certain candidates into particular positions. Thus, when the eigenvectors are lifted onto the permutahedron and the inner product is taken with the signal, it reveals how similar the data is to the given frame vector, which is interpretable.
(One minute is not long enough to capture the nice-ness of this, so please check out the full explanation on the mathematical walkthrough!)