Learning with Drift Detection

By João Gama, Pedro Medas, Gladys Castillo, Pedro Pereira Rodrigues. 2004

1,678 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Statistical Learning Theory

How researchers see this paper

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

In Artificial Intelligence

In Statistical Learning Theory

Citations in the corpus

Concrete Problems in AI Safety (2016, background): “There are many other ways to formalize this problem (for instance, in an online learning setting with concept drift [68, 52]) but we will focus on the above for simplicity.”
An Overview of Concept Drift Applications (2016, neutral): “[26]).”
Online Machine Learning in Big Data Streams (2018, background): “are diluted due to the new incoming data, which is more suitable for gradual changes [162, 73]. For sudden changes, algorithms that maintain a sliding window of the last seen instances perform better [228, 93, 145]. Another option is to include explicit forgetting mechanisms [142, 136, 76]. The most”
Early Drift Detection Method (2005, methodology): “The evaluation is similar to the one proposed in [2].”
Scikit-Multiflow: A Multi-output Streaming Framework (2018, neutral): “ti-output Learner ✓ ✓ ✓ ✓ ✓ * ✓ ✓ ✓ ✓ Bishop (2006) SGD ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Bishop (2006) Naive Bayes ✓ ✓ ✓ ✓ ✓ ✓ ✓ Bishop (2006) MLP ✓ ✓ ✓ ✓ ✓ ✓ ✓ Bishop (2006) ADWIN ✓ ✓ ✓ Bifet et al. (2018) DDM ✓ ✓ ✓ Gama et al. (2004) EDDM ✓ ✓ ✓ Bifet et al. (2018) Page Hinkley ✓ ✓ ✓ Page (1954) * Depending on the ”
From Concept Drift to Model Degradation: An Overview on Performance-Aware Drift Detectors (2022, methodology): “The Drift Detection Method (DDM) [30] is a well-known and widely-used algorithm and has been used as conceptual underpinning for a number of related performance-based drift detectors.”
Adaptive Anomaly Detection via Self-calibration and Dynamic Updating (2009, background): “Most publications which propose updating the model after significant changes to the environment, data stream, or application use supervised learning techniques, such as [12].”
The Impact of Diversity on Online Ensemble Learning in the Presence of Concept Drift (2010, background): “For example, in [14], several artificial data sets are considered to contain abrupt or gradual drifts.”
Reacting to Different Types of Concept Drift: The Accuracy Updated Ensemble Algorithm (2014, methodology): “The DDM algorithm and the windowed Hoeffding Tree were chosen as representatives of single classifiers .”
Data-driven Soft Sensors in the process industry (2009, background): “For detailed treatment and some solutions see Widmer and Kubat (1996); Gama et al. (2004).”
Online and Non-Parametric Drift Detection Methods Based on Hoeffding’s Bounds (2015, methodology): “Other theoretical results [12] have assumed that examples are generated independently and randomly from a sequence Pi of joint distributions over A! f0; 1g supposing that consecutive pairs of distributions have at most g total variation distance (0 < g 1).”
Issues in evaluation of stream learning algorithms (2009, background): “Several tests for change detection have been presented in the literature [2, 28, 22, 23, 13].”
Incremental on-line learning: A review and comparison of state of the art algorithms (2018, methodology): “It might be advisable to use dedicated techniques developed for non-stationary environments in applications where strong drift is expected, such as those incorporating explicit drift detection [86, 74] or recent approaches incorporating dedicated memory models [87].”
Classifier Ensembles for Detecting Concept Change in Streaming Data: Overview and Perspectives (2008, methodology): “In the classifier-free case any classifier can be used because the change detection and the window update depend only on the accuracy of the classification decision, and not on the model [10,20].”
Data stream clustering: A survey (2013, background): “On the other hand, a large window is desirable in stable phases, though it may not respond rapidly to concept drifts [Gama et al. 2004].”