China 
Africa 
Explore chapters and articles related to this topic
k-Means Clustering
Published in Mohssen Mohammed, Muhammad Badruddin Khan, Eihab Bashier Mohammed Bashier, Machine Learning, 2016
Mohssen Mohammed, Muhammad Badruddin Khan, Eihab Bashier Mohammed Bashier
The algorithm has a loose relationship to the k-nearest neighbors (k-NN) classifier, a popular machine learning technique for classification that is often confused with k-means because of the k in the name. One can apply the 1-nearest neighbor classifier on the cluster centers obtained by k-means to classify new data into the existing clusters. This is known as nearest centroid classifier or Rocchio algorithm.
Experiments in cross-domain few-shot learning for image classification
Published in Journal of the Royal Society of New Zealand, 2023
Hongyu Wang, Henry Gouk, Huon Fraser, Eibe Frank, Bernhard Pfahringer, Michael Mayo, Geoffrey Holmes
Prototypical networks (Snell et al. 2017) are the most popular example of episodic training in few-shot learning. In each episode of training, a nearest centroid classifier is constructed, and the cross-entropy loss of this classifier is used as a training signal for the feature extractor. It is instructive to relate this to the classification methods we consider in this paper. Nearest centroid classifiers can be seen as a simplification of linear discriminant analysis (LDA) (Hastie et al. 2009). While nearest centroid classifiers commonly use Euclidean distance, LDA makes use of Mahalanobis distance – a metric defined via the covariance matrix of the training data. The quadratic discriminant analysis (QDA) extension constructs a per-class covariance matrix, rather than a global pooled covariance matrix. We also consider the shrunken nearest centroid classifier (Tibshirani et al. 2003), a variant of LDA designed for high-dimensional datasets with only a few samples – a defining characteristic of few-shot learning. Finally, addressing high-dimensional problems with LDA, it is common to construct a committee of classifiers, where each member of the ensemble first randomly projects the high-dimensional features into a lower-dimensional space (Durrant and Kabán 2015). We include all these variants in our experiments.