Interesting result. Even more interesting that it’s come out of AOL. I figured AOL was dead – not the case. They’ve pivoted into adtech. Nice that they’re publishing some of their work!

HLL was my absolute favorite algorithm for a little while after I discovered it. I had never really seen the power of probabilistic algorithms before, but understanding it (and implementing it) really opened my eyes. Nice to see improvements being made on it.

I wonder if someone in the future will be able to make some mathematical proofs about the beta function and the form of an optimal one, though. The gist of this paper is basically just “we realized that you could probably unbias HLL using a continuous function instead of changing regime like HLL++, so we did a bunch of curve-fitting and… it works!” It’s a good empirical result, but there’s no proof element to it.

Interesting result. Even more interesting that it’s come out of AOL. I figured AOL was dead – not the case. They’ve pivoted into adtech. Nice that they’re publishing some of their work!

HLL was my absolute favorite algorithm for a little while after I discovered it. I had never really seen the power of probabilistic algorithms before, but understanding it (and implementing it) really opened my eyes. Nice to see improvements being made on it.

I wonder if someone in the future will be able to make some mathematical proofs about the beta function and the form of an optimal one, though. The gist of this paper is basically just “we realized that you could probably unbias HLL using a continuous function instead of changing regime like HLL++, so we did a bunch of curve-fitting and… it works!” It’s a good empirical result, but there’s no proof element to it.