The problem of concept drift: definitions and related work

By Alexey Tsymbal. 2004

1,184 citations recorded by the source corpus.

Subfields: Meta & Continual Learning, Time-Series & Forecasting

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Meta & Continual Learning

In Time-Series & Forecasting

Citations in the corpus

Ups and Downs: Modeling the Visual Evolution of Fashion Trends with One-Class Collaborative Filtering (2016, methodology): “; see the work of Tsymbal [34] for a comprehensive survey.”
From amateurs to connoisseurs: modeling the evolution of user expertise through online reviews (2013, background): “Some early works that deal with temporal dynamics do so in terms of concept drift [23, 37, 39].”
Continual Prototype Evolution: Learning Online from Non-Stationary Data Streams (2020, background): “Besides the focus on continual learning in this work, our learner-evaluator framework generalizes to concept drift as well [38], for which we refer to an overview in [45, 12].”
Ensemble learning for data stream analysis: A survey (2017, background): “For more information on these and other changes in underlying data distributions, the reader is referred to [62, 65, 116, 184, 176].”
A Review on Outlier/Anomaly Detection in Time Series Data (2020, methodology): “(2017) propose two algorithms based on this theory: SPOT, for data following any stationary distribution; and DSPOT, for data that can be subject to concept drift (Tsymbal, 2004; Gama et al., 2014).”
Detecting concept drift in data streams using model explanation (2018, background): “, 2014), even though the data distributions and concepts underlying those datasets tend not to remain stable over a longer period of time (Widmer & Kubat, 1996; Tsymbal, 2004).”
The CART decision tree for mining data streams (2014, background): “Additionally, the concept drift may occur [8] [10] [17] [20] [22] [34], which means that the concept of data evolve in time.”
Incremental learning algorithms and applications (2016, background): “Changes in the data distribution over time are commonly referred to as concept drift [88, 157, 33, 126].”
Characterizing concept drift (2015, background): “The concept of abrupt drift [Tsymbal, 2004, Hoens et al., 2012, Zliobaite, 2010, Bose et al., 2011, Huang et al., 2013, Minku et al., 2010, Dongre and Malik, 2014, Brzezi´nski, 2010, Brzezinski and Stefanowski, 2014a] illustrates this implicit assumption.”
Reacting to Different Types of Concept Drift: The Accuracy Updated Ensemble Algorithm (2014, background): “to [1], [5], [7], [17], [18] for more information on class lab el swaps or changes in underlying data distributions.”
Review of soft sensor methods for regression applications (2016, methodology): “with multiple concept descriptors) [113].”
Incremental learning with self-organizing maps (2017, background): “Generally, changes in data statistics as encountered in incremental learning are denoted somewhat generally as concept drift [9], [10], which can be gradual or abrupt.”
A Review of Challenges and Opportunities in Machine Learning for Health. (2018, methodology): “In developing new models for healthcare, models must be made robust to these changes, or acknowledge their mis-calibration for the new population 79 .”
Incremental Clustering for the Classification of Concept-Drifting Data Streams (2008, background): “As other researchers have noted [17], methodologies proposed for tackling with concept drift can be divided in three main groups.”
Continuous Domain Adaptation with Variational Domain-Agnostic Feature Replay (2020, background): “, virtual concept drift (Widmer & Kubat, 1993; Tsymbal, 2004), feature change (Gao et al.”
Enabling Fast Lazy Learning for Data Streams (2011, neutral): “[15] A.”
A comprehensive review of digital twin — part 1: modeling and twinning enabling technologies (2022, background): “Such drift has led to substantial ML model performance degradation (Tsymbal, 2004).”
Machine Learning With Big Data: Challenges and Approaches (2017, methodology): “learning models are built using older data that no longer accurately reflect the distribution of new data [55].”