Determine the mechanism driving MSP-score ageing

Determine the mechanism responsible for the partial narrowing of the teacher–student unknown-detection gap under the maximum softmax probability score, beyond the contribution attributable to ensemble aggregation.

Background

The study finds that the teacher–student unknown-detection gap narrows over time rather than widens. For the energy score, ensemble aggregation does not explain the result because the ensemble and its individual members age at essentially the same rate. For the maximum softmax probability score, however, aggregation explains approximately one quarter of the narrowing, leaving the remainder unexplained. The authors therefore identify the underlying mechanism for the MSP result as unresolved.

References

Aggregation accounts for about a quarter of the MSP narrowing rather than all of it, so it is a partial explanation there and not a sufficient one. The mechanism is settled for one co-primary score and partly open for the other.

— Unknown-Traffic Detection, Calibration and Shortcut Reliance in Distilled Encrypted-Traffic Classifiers over One Year  (2609.31141 - Abbasi, 25 Sep 2026) in Section 5.5, “Aggregation as an Explanation for the Ageing”