NeurIPS paper

[lacerbi]

• Write Methods and sketch Intro/Background
• Write sketch of abstract

[lacerbi]

8/5/2023 1:15PM

I wrote the abstract and most sections of the paper on Overleaf (except for the Experiments of course, where @Grégoire Clarté added a couple of stubs). If you make any changes (unless it's typos or super-minor stuff), please do so in revision mode, or add a comment, otherwise it becomes a mess.

General status:

• All: Have a look at the abstract (still a draft), since we need to submit in a couple of days.
• The Introduction is a bit weak and quite draft-y; we will strengthen it once we know better our actual contributions and what ends up being in the paper.
• Background, RCNP and Proof look reasonably solid to me, but feedback is welcome! A few pieces here and there are missing but should be easy to fill in. I plan to polish it in the future, feel free to add stuff (in edit mode). Of course we can also trim some stuff later.

What's missing:

• Obviously Experiments. It'd be great if each of you can start to fill in a brief description of their section. Assume that you have less than a page each in total for your section (including figures and results), more likely 2/3 of a page.
• Discussion should be easy to fill in quickly, after we have the results.
• Related work: it'd be good if someone with a bit of experience in the area (metalearning etc.) can start filling it in, at least with a few stubs. This should be short.
• Background: It'd be great if someone can take the time to draft the final part of the section, especially the ConvCNP paragraph, with proper notation consistent with our paper. For the other (C)NP models, like the LNPs, AttnCNPs, AR-CNP, I was just planning to list them with a very brief description (one or two sentences each), but not the full definition.
• In the RCNP section we could have more stuff, like how we handle multi-output stuff, etc., but it looks like it's Appendix material. One important thing which is missing (since it's unclear if we will implement it) is the permutation equivariance with respect to input features.
• Anything else major which is missing?

Of course, give absolute priority to working on the results; but if you are just sitting around waiting for a run to finish, contributions to writing are welcome.