Drift Localization using Conformal Predictions

arXiv:2602.19790v2 Announce Type: replace Abstract: Concept drift -- the change of the distribution over time -- poses significant challenges for learning systems and is of central interest for monitoring. Understanding drift is thus paramount, and drift localization -- determining which samples are affected by the drift -- is essential. While several approaches exist, most rely on local testing schemes, which tend to fail in high-dimensional, low-signal settings. In this work, we consider a fundamentally different approach based on conformal predictions. We discuss and show the shortcomings of common approaches and demonstrate the performance of our approach on state-of-the-art image datasets.

Related

• On the Theory of Continual Learning with Gradient Descent for Neural Networks
• From Order to Distribution: A Spectral Characterization of Forgetting in Continual Learning
• eCP: Equivariant Conformal Prediction with pre-trained models

Source: arXiv cs.LG | 2026-04-22