Feature Shift
4896·updated Aug 28, 2026Unlike 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.
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
Sensitivity
Information Class
Variants
- plural
- Feature Shifts
- possessive
- Feature Shift's
- pluralpossessive
- Feature Shifts'
Framework definitions
- §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.
- §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