2nn

2nn is an abbreviation that can refer to two-nearest-neighbor concepts in various fields, most notably in machine learning and pattern recognition. In common usage within this domain, 2NN denotes the two-nearest-neighbor classifier, a simple variant of the k-nearest-neighbors (kNN) algorithm used for classification and, less commonly, regression. In a 2NN classifier, the label of a new instance is determined by the labels of its two closest training samples, measured by a chosen distance metric such as Euclidean distance.

The typical decision rule is straightforward: if the two nearest neighbors share the same class, that class is assigned to the new instance; if they disagree, a tie-breaking rule is applied. Common approaches include selecting the class of the nearer neighbor or using distance-weighted voting, where closer neighbors have more influence. 2NN is a non-parametric method and often serves as a baseline due to its simplicity, but its performance can be sensitive to noise, outliers, and the chosen distance metric. Computationally, prediction can be linear in the number of stored samples unless a data structure or indexing method is used to speed up neighbor searches.

In addition to classification, 2NN may appear in other contexts. Some literature uses 2NN to denote a two-layer neural network, referring to a shallow neural network with a single hidden layer. Context is important to determine the intended meaning, as 2NN can also arise in discussions of two-nearest-neighbor graphs or related constructs in computational geometry.