Abstract

As statistical data sets grow larger and larger, the availability of fast and efficient algorithms becomes ever more important in practice. Classical methods are often easy to compute, even in high dimensions, but they are sensitive to outlying data points. Robust statistics develops methods that are less influenced by abnormal observations, often at the cost of higher computational complexity. Many robust methods, especially those based on ranks, are closely related to geometric or combinatorial problems. An early overview of relations between statistics and geometry was given in [Sha76].