Why it matters

Frequency estimation at stream scale requires approximation. Count-Min is standard approach.

Advertisement

The architecture

Matrix of d × w counters. d hash functions.

Increment: hash key with all d; increment each corresponding counter.

Query: hash key; return min of d counters.

Count-Min SketchMatrix d × wcountersUpdateincrement d cellsQuerymin of d cellsError bound: over-estimate at most ε × total count with probability 1-δ; tune d, w
Sketch operation.
Advertisement

How it works end to end

Error bound: ε × total count with probability 1-δ. Tune d, w for target.

Heavy hitters: track items with highest estimated counts.

Applications: network measurement, database query optimization, streaming ML.

Relates to Bloom filter but for counts.