Catching Every Ripple: Enhanced Anomaly Awareness via Dynamic Concept Adaptation

arXiv cs.LG / 4/17/2026

📰 NewsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The article introduces DyMETER, a dynamic concept adaptation framework aimed at improving online anomaly detection under concept drift without frequent retraining.
DyMETER combines on-the-fly parameter shifting and dynamic thresholding in a single online paradigm, transitioning from a static detector to a dynamic mode when drift is detected.
It uses a hypernetwork to produce instance-aware parameter shifts, allowing efficient adaptation while avoiding costly fine-tuning or retraining.
The method adds a lightweight evolution controller to estimate instance-level concept uncertainty and a dynamic threshold optimization module that recalibrates the decision boundary using a candidate window of uncertain samples.
Experiments on multiple scenarios reportedly show DyMETER substantially outperforms existing online anomaly detection approaches, with an emphasis on robustness and interpretability.
The work is presented as a new arXiv submission (arXiv:2604.14726v1), signaling an early-stage research contribution rather than a deployed product update.

Abstract

Online anomaly detection (OAD) plays a pivotal role in real-time analytics and decision-making for evolving data streams. However, existing methods often rely on costly retraining and rigid decision boundaries, limiting their ability to adapt both effectively and efficiently to concept drift in dynamic environments. To address these challenges, we propose DyMETER, a dynamic concept adaptation framework for OAD that unifies on-the-fly parameter shifting and dynamic thresholding within a single online paradigm. DyMETER first learns a static detector on historical data to capture recurring central concepts, and then transitions to a dynamic mode to adapt to new concepts as drift occurs. Specifically, DyMETER employs a novel dynamic concept adaptation mechanism that leverages a hypernetwork to generate instance-aware parameter shifts for the static detector, thereby enabling efficient and effective adaptation without retraining or fine-tuning. To achieve robust and interpretable adaptation, DyMETER introduces a lightweight evolution controller to estimate instance-level concept uncertainty for adaptive updates. Further, DyMETER employs a dynamic threshold optimization module to adaptively recalibrates the decision boundary by maintaining a candidate window of uncertain samples, which ensures continuous alignment with evolving concepts. Extensive experiments demonstrate that DyMETER significantly outperforms existing OAD approaches across a wide spectrum of application scenarios.