[D] How ICLR review scores changed during the discussion period
I have gathered some statistics about the publicly available ICLR review data. In particular, I was interested in how the scores were affected by the discussions.
In summary, 11.79% of all reviews changed their score (mostly improvements).
Surprisingly, reviewers who stated that they “I do not know much about this area” or “I made a quick assessment of this paper” were less likely to change their score.
Detailed results and code can be found here