News for hackers. Stuff that matters.
Re:Billion-Scale Approximate Nearest Neighbor Search [pdf] by nwoli on Saturday May 06, 2023 @04:31PM (#35845118)
Spotify uses random hyperplanes to produce a hash where each bit is the sign of the dot product of the query and the hyperplanes (ie “is above or below” the hyperplane). Works really well
Re:Billion-Scale Approximate Nearest Neighbor Search [pdf] by AndrewKemendo on Saturday May 06, 2023 @06:32PM (#35845926)
Is there a blog post or something on this?
Re:Billion-Scale Approximate Nearest Neighbor Search [pdf] by gregw134 on Saturday May 06, 2023 @08:04PM (#35846556)
I think that's just a variation on locality sensitive hashing. You can find lots of articles online.
Re:Billion-Scale Approximate Nearest Neighbor Search [pdf] by fancy_pantser on Saturday May 06, 2023 @08:56PM (#35846809)
There's at least a repo: https://github.com/spotify/annoy
Re:Billion-Scale Approximate Nearest Neighbor Search [pdf] by AndrewKemendo on Saturday May 06, 2023 @09:17PM (#35846926)
Thank you!
Re:Billion-Scale Approximate Nearest Neighbor Search [pdf] by ChrisPzilla on Tuesday May 09, 2023 @11:53AM (#35876273)
Here is an article on Indexes https://zilliz.com/blog/vector-index
Re:Billion-Scale Approximate Nearest Neighbor Search [pdf] by nwoli on Wednesday May 10, 2023 @10:53AM (#35888256)
It’s wrong though, annoy is not tree based