Really funny that you mention that! Some of our more recent work focuses on using adaptive sampling techniques in approximate-nearest-neighbor search (actually, the related problem of maximum inner product search: https://arxiv.org/abs/2212.07551).

We definitely think that our approach could be used to make an index structure for ANN search directly, for example in conjunction with Hierarchical Navigable Small World approaches.

Thanks for sharing your great work!

[Try this tool from arxiv labs (just replace the x in original url with a 5)]

https://ar5iv.labs.arxiv.org/html/2006.06856

https://ar5iv.labs.arxiv.org/html/2212.07551

Oh cool, neat trick, thanks!