home/glossary/Feature Shift

Feature Shift

nounid 4896·updated Aug 28, 2026
verified

Unlike joint distribution shift detection, which cannot localize which features caused the shift, we define a new hypothesis test for each feature individually. Naïvely, the simplest test would be to check if the marginal distributions have changed for each feature (as explored by [25]); however, the marginal distribution would be easy for an adversary to simulate (e.g., by looping the sensor values from a previous day). Thus, marginal tests are not sufficient for our purpose. Therefore, we propose to use conditional distribution tests. More formally, our null and alternative hypothesis for the j-th feature is that its full conditional distribution (i.e., its distribution given all other features) has not shifted for all values of the other features.

MWE

Attested in

No recorded attestations. They are written when an MWE tagging stage is completed, stamped with the pack version and the document’s digest.

Classifications

Entity Type

Metric60%llm-generatedllm:claude-haiku-4-5

Sensitivity

unclassified

Information Class

unclassified

Variants

plural
Feature Shifts
possessive
Feature Shift's
pluralpossessive
Feature Shifts'

Framework definitions

Feature Shift Detection: Localizing Which Features Have Shifted via Conditional Distribution Tests1 senseview framework →
§1
Unlike joint distribution shift detection, which cannot localize which features caused the shift, we define a new hypothesis test for each feature individually. Naïvely, the simplest test would be to check if the marginal distributions have changed for each feature (as explored by [25]); however, the marginal distribution would be easy for an adversary to simulate (e.g., by looping the sensor values from a previous day). Thus, marginal tests are not sufficient for our purpose. Therefore, we propose to use conditional distribution tests. More formally, our null and alternative hypothesis for the j-th feature is that its full conditional distribution (i.e., its distribution given all other features) has not shifted for all values of the other features.
Legacy lexicon import1 senseview framework →
§1 · legacy_primary
Unlike joint distribution shift detection, which cannot localize which features caused the shift, we define a new hypothesis test for each feature individually. Naïvely, the simplest test would be to check if the marginal distributions have changed for each feature (as explored by [25]); however, the marginal distribution would be easy for an adversary to simulate (e.g., by looping the sensor values from a previous day). Thus, marginal tests are not sufficient for our purpose. Therefore, we propose to use conditional distribution tests. More formally, our null and alternative hypothesis for the j-th feature is that its full conditional distribution (i.e., its distribution given all other features) has not shifted for all values of the other features.
DR-088 backfill from the noun definition column

Outgoing relationships

No outgoing triples
This term is not the subject of any RDF-style relationship yet.

Incoming relationships

No incoming triples
No other term currently asserts a relationship to this one.