Drift Detection
Drift detection focuses on identifying changes in the statistical properties of data streams over time, a phenomenon known as concept drift, impacting the performance of machine learning models. Current research emphasizes developing unsupervised methods, particularly those leveraging deep learning representations (like autoencoders) and generative adversarial networks (GANs), to detect both global and localized drifts, as well as addressing the challenges posed by high-dimensional data and limited labeled data. Effective drift detection is crucial for maintaining the accuracy and reliability of machine learning systems across diverse applications, from wildlife monitoring and manufacturing to fraud detection and personalized recommendations, enabling timely model adaptation and improved system performance.
Papers
Localizing Anomalies in Critical Infrastructure using Model-Based Drift Explanations
Valerie Vaquet, Fabian Hinder, Jonas Vaquet, Kathrin Lammers, Lars Quakernack, Barbara Hammer
One or Two Things We know about Concept Drift -- A Survey on Monitoring Evolving Environments
Fabian Hinder, Valerie Vaquet, Barbara Hammer
Autoencoder-based Anomaly Detection in Streaming Data with Incremental Learning and Concept Drift Adaptation
Jin Li, Kleanthis Malialis, Marios M. Polycarpou
DA-LSTM: A Dynamic Drift-Adaptive Learning Framework for Interval Load Forecasting with LSTM Networks
Firas Bayram, Phil Aupke, Bestoun S. Ahmed, Andreas Kassler, Andreas Theocharis, Jonas Forsman