Autoassigner support for PC auto-division

[kohler]

From Dan Tsafrir and Emmett Witchel:

I'm not sure HotCRP can (or should) help program chairs with the second problem I'm going to describe, but I'm going to describe it nevertheless just in case. The most challenging issue of running a dual track meeting, I'm told, is deciding which heavy program members will review which papers, such that a split into two rooms is possible/manageable. In ASPLOS'19, the approach taken by Emmett (cc-ed) is partitioning the heavy PC members into two disjoint sets of similar size:

H1 \union H2 = HeavyPC; H1 \intersection H2 = \emptyset; |H1| =~ |H2|

and partitioning the submitted papers into two disjoint sets of similar size:

P1 \union P2 = Papers; P1 \intersection P2 = \emptyset; |P1| =~ |P2|

such that H1 members are assigned to review papers form only P1, and H2 members are assigned to review papers form only P2, and the assignment optimizes reviewers' "satisfaction" based on their bids and topics of interest. This is a (NP) hard problem -- Emmett tells me that he used some script that executed for a good few days after the submission deadline until producing a reasonable assignment,…

[dants]

Please see more information about this issue and how we eventually addressed it in Section 6 of

https://www.usenix.org/sites/default/files/atc19_message.pdf