Add support for multiple claims per paper

robust-ml / robust-ml.github.io

A community-run reference for state-of-the-art adversarial example defenses.

https://www.robust-ml.org/

Creative Commons Attribution Share Alike 4.0 International

49 stars 7 forks source link

Add support for multiple claims per paper #8

Closed anishathalye closed 5 years ago

anishathalye commented 5 years ago

As the title says.

This change gets rid of sorting on threat model / dataset (it wasn't all that useful anyways, in my opinion; filtering still works). It also adds a sticky header.

This change also includes an example of multiple claims per paper (from #4).

cc @max-andr

anishathalye commented 5 years ago

And here is a screenshot, so people don't have to run this themselves to see what it looks like:

max-andr commented 5 years ago

Looks nice in my opinion. Maybe adding a bit more horizontal space to column "Claims" would improve the readability of the table.

One small thing regarding our numbers (from the screenshot) - on MNIST we have the results not on 1k, but on 10k points (full test set). And the empirical adversarial accuracy is 96.42% instead of 96.4%. Just to be precise in the claims :)

anishathalye commented 5 years ago

Oops, must have missed that. Yes, we certainly care about being precise in the claims -- fixed :)

I tried playing with the column spacing, but I couldn't get it to look much better. I'm going to merge this as-is, and if anyone can figure out how to make the table layout better, then a separate pull request would be appreciated.