intermediate

Streaming & Online Machine Learning

Machine learning on unbounded data — the one-pass streaming model, sublinear sketches (reservoir sampling, Bloom filters, Flajolet-Martin, DGIM), online learning and regret, concept drift detection, and production streaming architectures.

The Streaming ModelEvent Time vs Processing TimeReservoir SamplingBloom FiltersFlajolet-Martin & Distinct CountsDGIM & Sliding WindowsOnline Gradient DescentRegret & Online-to-BatchFollow-the-Regularized-LeaderConcept Drift DetectionADWIN & DDMHoeffding TreesFeature Freshness & Online/Offline ParityStreaming Architectures
ML project loop3/6 stages →
DataHypothesis spaceObjectiveOptimizationEvaluationDeployment feedback