Title: Extreme value theory for open set classification: The GPD and GEV classifiers
Authors: Edoardo Vignotto - University of Geneva (Switzerland) [presenting]
Sebastian Engelke - Ecole Polytechnique Federale de Lausanne (Switzerland)
Abstract: Classification tasks usually assume that all possible classes are present during the training phase. This is restrictive if the algorithm is used over a long time and possibly encounters samples from unknown classes. It is therefore fundamental to develop algorithms able to distinguish between known and unknown new data. In the last few years, extreme value theory has become an important tool in multivariate statistics and machine learning. The recently introduced extreme value machine, a classifier motivated by extreme value theory, addresses this problem and achieves competitive performance in specific cases. However, this algorithm can fail when the geometries of known and unknown classes differ, even if the recognition task is fairly simple. To overcome these limitations, two new algorithms for open set classification relying on approximations from extreme value theory that are more robust in such cases are proposed. They exploit the intuition that test points that are extremely far from the training classes are more likely to be unknown objects. Asymptotic results motivated by univariate extreme value theory that make this intuition precise are proposed. The effectiveness of the new classifiers is shown in simulations and on real data sets.