DOC3: Deep One Class Classification Using Contradictions

Abstract

This paper introduces the notion of learning from contradictions (a.k.a Universum learning) for deep one class classification problems. We formalize this notion for the widely adopted one class large-margin loss (Schölkopf et al. in Neural Comput 13(7):1443–1471), and propose the deep one class classification using contradictions (DOC $^3$ 3 ) algorithm. We show that learning from contradictions incurs lower generalization error by comparing the empirical Rademacher complexity of DOC $^3$ 3 against its traditional inductive learning counterpart. Further, our proposed ‘learning from contradiction’ is a generic learning setting and can compliment other advanced learning settings. To illustrate this, we extend the adversarial learning based DROCC-LF (Goyal et al. in International conference on machine learning, PMLR, 2020) algorithm under this new setting. Our empirical results demonstrate the efficacy of DOC $^3$ 3 and it’s extensions compared to popular baseline algorithms on several benchmark and real-life data sets.