This was a variation of the 2011 PFAC talk to introduce Flajolet’s streaming legacy, given at a conference dedicated to bridging the gap between implementers and researchers, organized by Aggregate Knowledge.
As a PhD student at Inria Rocquencourt, I gave a talk that provided an overview of how data streaming algorithms like HyperLogLog work, that was aimed at a broader audience of engineering PhD students at the Inria’s Junior Seminar.
Talk that introduces a novel, implementable data streaming algorithm for cardinality estimation, that uses
This talk traces the history of data streaming algorithms for the cardinality estimation problem from the 1980s to 2010s through algorithms designed or studied by Philippe Flajolet, and the origins of the world-famous HyperLogLog algorithm.
Talk that introduced a novel, practically useful cardinality estimation, which is both optimal and fully analyzed, engineered with low cardinality distortion fixes, with its limit distribution (see sequence of plots): A full characterization of the probability distribution — a gamma distribution — of the algorithm’s output.