[The NeurIPS way] A2+B1

[mavroudisv]

IMHO the best options are A2 and B1 ie the way NeurIPS does it.

• Having talks last less than two minutes provides very little information, it's a logistic nightmare (as pointed out in another issue) and will waste at least as much time in-between the talk.

• Having talks getting randomly selected seems suboptimal when we have the review scores available. In case these scores are too coarse to distinguish the top 20 (percentages can complicate things as the submission numbers grow) or so papers, another lightweight metric could be introduced (e.g., each reviewer ranks their accepted papers in order of preference for presentation/spotlight).

• This approach is already tested and designed to handle a very large number of papers (ie the reason we are considering changing the presentation mode).