Learning the k in k-means

By Greg Hamerly, Charles Peter Elkan. Neural Information Processing Systems, 2003

1,180 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/2555258da9bdc242fe0681b0d6d6a76db912a098

How researchers see this paper

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

Citations in the corpus

Application and Evaluation of a Hierarchical Patch Clustering Method for Remote Sensing Images (2016, methodology): “Based on the method proposed in [17], the Anderson–Darling Gaussian test performs better than the Bayesian information criterion (BIC) in finding a good stopping rule during clustering.”
Instance-Specific Algorithm Configuration (2014, methodology): “The previous chapters used g-means[37] to analyze the training data and find a stable set of clusters.”
Efficient and Robust Automated Machine Learning (2015, unknown): “Specifically, we clustered our 140 datasets with g-means [27] based on the dataset meta-features and used one dataset from each of the resulting 13 clusters (see Table 6 in the supplementary material for the list of datasets).”
Efﬁcient and Robust Automated Machine Learning (2015, unknown): “Specifically, we clustered our 140 datasets with g-means [27] based on the dataset meta-features and used one dataset from each of the resulting 13 clusters (see Table 6 in the supplementary material for the list of datasets).”
Dynamic clustering using particle swarm optimization with application in image segmentation (2006, methodology): “the number of clusters within a data set, in advance, is a difficult problem by itself [10].”
Transferable AutoML by Model Sharing Over Grouped Datasets (2019, methodology): “Statistical test was involved in k-means by [13], whereby the test is performed to detect whether the data assigned to a cluster is sampled from one Gaussian distribution.”
Revisiting the Impact of Classification Techniques on the Performance of Defect Prediction Models (2015, methodology): “K-means divides the data into k clusters and centroids are chosen randomly in an iterative manner [22].”
Taking Human out of Learning Applications: A Survey on Automated Machine Learning (2018, methodology): “These motivates many tailormade methods for the selection of various machine learning models, such as the kernel selection for SVM [143], learning K for K-means [144] and genetic programming for neural networks architectures [80], [82].”
Automated Algorithm Selection: Survey and Perspectives (2018, background): “over all solvers from a given set, Kadioglu et al. (2011) devised a different approach, in which they partitioned the (normalised) instance feature space for a given training set by means of g-means (Hamerly and Elkan, 2003). Then, the best value of k to be used by the k-nearest neighbour algorithm ”
Automated Machine Learning: Methods, Systems, Challenges (2019, unknown): “More complete surveys can be found in the literature [26, 98, 130, 138, 175].”
Exploratory Learning (2013, background): “There has also been substantial work in the past to automatically decide the right “number of clusters” in unsupervised learning [11,22,15,7,19,27].”
On learning and branching: a survey (2017, methodology): “The grouping of instances is carried out by the g-means algorithm of Hamerly and Elkan (2003), a method similar to the classical k-means of MacQueen (1967), which additionally determines in an automatic way the optimal number of clusters, assuming an underlying Gaussian density distribution.”
Revisiting k-means: New Algorithms via Bayesian Nonparametrics (2011, methodology): “s assume a ﬁxed, known value for kwhich, in many cases, is simply not known a priori. Some work has gone into the development of techniques to overcome this problem in the context of k-means, such as [2, 7, 8, 12, 14, 19, 21], which utilize various model selection criteria. In the case of mixture mo”
Machine Learning in Aerodynamic Shape Optimization (2022, background): “In addition, the cluster number needs to be selected beforehand, leading to a difficult decision [192].”
K-means clustering: a half-century synthesis. (2006, methodology): “Hamerly and Elkan (2002) created an algorithm, G-means, to determine K by choosing a small value for K, running the K-means algorithm, and testing whether the data assigned to each cluster centre are from a Gaussian distribution.”