From outliers to prototypes: Ordering data

摘要: We propose simple and fast methods based on nearest neighbors that order objects from high-dimensional data sets typical points to untypical points. On the one hand, we show these easy-to-compute orderings allow us detect outliers (i.e. very points) with a performance comparable or better than other often much more sophisticated methods. how use prototypes (very which facilitate exploratory analysis algorithms such as noisy nonlinear dimensionality reduction clustering. Comprehensive experiments demonstrate validity of our approach.