AI Glossary

Distribution shift

What does “Distribution shift” mean?

Distribution shift: A change between the distribution a model learned from and the one it encounters later.

Distribution shift visual guide: A change between the distribution a model learned from and the one it encounters later.
Visual guide Open full size (new tab)

What is an example of “Distribution shift”?

A speech model trained on studio audio is used on phone calls.

Keep learning

See also

EvalsTermBenchmarkTermValidation and test setsTermStratified samplingTermData leakageTerm

Same topic

Concept driftTermCovariate shiftTermError analysisTermDifferential privacyTermEU AI ActTerm