Years ago, ICML decided to hold a special session of paper trashed by reviewers but rescued by ACs. (E.g. because of an interesting but unfashionable idea, and reviewers missed see the point) It was the most interesting session of the entire conference!https://twitter.com/yeewhye/status/1388244429694246914 …
Indeed. Or more precisely, high variance of reviews is a great heuristic for selecting papers. (And probability of rejection as a function of quality is bimodal: the truly groundbreaking papers are often rejected because the reviewers don't get them.)
