Optimize the CSranking of
Why did I do this?
I dunno. Twiddling the boxes on CSrankings felt meaningful? Now it doesn't? There are 45 institutions that are "Number One" as of 13-Jun-2021.
How did I do this?
This was fun! The csrankings code on github (https://github.com/emeryberger/CSRankings) will generate a table of adjusted counts per institution per venue (grouped into subfields). It determines the rank according to the geometric mean of this count (+1) for the selected areas. The geometric mean is the arithmetic mean in log space. The rank of an institution is the number of institutions with a higher geometric mean (plus one). This is the L0 norm of the rectified difference of means. Relaxing this to the L1 norm, I applied gradient descent from ~60 random starting positions, breaking ties to maximize count distance to the next best ranked institution. I bought a faster computer and now I'm just brute-forcing the 2^26 different topic-area checkbox combinations. The relaxation is probably still a good idea when gaming individual venues (2^75), but CSrankings URLs can encode topic areas but not venue selections.
Does this make it easy to see the unique strengths of particular institutions?
In general, no. Often boxes are selected not because a school is particularly good at that field, but rather because selecting it more negatively affects competitors. For example, Univeristy of Toronto is CSranked 3rd in the world for HCI and 5th for Graphics. With ~14 faculty in these areas, its safe to say University of Toronto is "good" at HCI and Graphics. If we try to game CSrankings' checkboxes by checking HCI and Graphics then this puts University of Toronto only in 2nd CSrank: below CMU. To game all the way to #1, Visualization needs to be included. CSranking on just Visualization, University of Toronto is merely 23rd. Importantly however, CMU is significantly lower at 173th. If you're a student interested in HCI, Graphics and Visualization, then checking these boxes could still be meaningful. But if you're a department chair, hiring committee member or faculty crawling CSrankings to size up your department relative to others, then this emergent behavior in these gaming results should give you great pause to reflect on whether your endeavour is worthy. I've created another site, CSRankings Profile, which generates a more wholistic view of an institution based in its csrankings.