search

semitrivial / AgentMixtures

LaTeX/Code for "Agent mixtures and the genericness of non-deterministic intelligence"
0 stars 0 forks source link

Pad the deleted discussion about previous paper as part of the conclusion? #27

Open lddllddl opened 1 year ago

lddllddl commented 1 year ago

I think it can be trimmed down to be both shorter and "softer" and suit the Conclusion very well, also adding back a few references.

lddllddl commented 1 year ago

Also currently we are at 16 pages. I should be able to add further 1 to 1.5 pages' worth of content as more discussion in Conclusion and additional references before deadline. So feel free to add back more proofs in Supplementary Materials.

semitrivial commented 1 year ago

I don't think we should intentionally pad the paper just to increase the page-count. Besides, there's an 8-page limit for content other than bibliography/appendix, isn't there?

lddllddl commented 1 year ago

FYI I have uploaded the 2022 theory papers in a branch. https://github.com/semitrivial/AgentMixtures/pull/33/files (quite a few are not really theoretic ones - but I kept them to avoid further bias with filtering one-by-one) No negative samples available, but I would personally predict increased page count to increase the chance of acceptance, by "unsupervised learning" the positive samples.

My speculation is that the reviewers tend to first assess the overall "fanciness" of the paper (which correlates very heavily with the length I suppose - COLT papers are clearly longer than ALT ones with the same format), and then (especially for the extra proofs in the appendix) only verify the technical correctness of some of the content.

So it would be a bad idea to generate new content just to pad the length, since they could very well be low-quality and reduce the expected perception of quality when the reviewer choose to examine a random part of the paper (including supplementary). But contents that we already consider to be of high quality, when added back, will increase the perceived overall fanciness, while maintaining the expected perception of quality when randomly sampled.

Considering the distribution of total lengths of published papers, I would estimate the marginal utility of increasing the page count in terms of increasing the overall "fanciness", when the page count is below 20 pages, to be significant.

lddllddl commented 1 year ago

That again, adding the conclusion and the fact that final versions usually increase in length, I'd say current length is in the safe range and not exactly in dire need of padding. 17 pages in final version is a common size in fact.