Definition
Plain language
A fast way to group incoming items on the fly by comparing each new one to a few running summaries instead of re-sorting everything.
As stated in the literature
An online clustering technique that maintains a small set of running cluster summaries (centroids) and assigns each arriving item to the nearest, avoiding the quadratic cost of re-clustering; the basis for scalable cross-conversation attack monitoring.
Why it matters: It groups a fast-arriving flow of data cheaply, making it practical to watch many conversations at once for coordinated attacks.
For example, as new customer reviews arrive, it sorts each one into an existing group by comparing it to a few running summaries instead of resorting every review.
Heard on the show
“And the trick that makes it possible is borrowed from an old corner of data mining — stream clustering.”Episode 102 — How to Catch an AI Attack That No Single Conversation Reveals