Learning from positive and unlabeled examples

By François Denis, Rémi Gilleron, Fabien Letouzey. Theoretical Computer Science, 2000

296 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/5b7addfb161b6e43937c9b8db3c85f10de671d0c

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Positive-Unlabeled Learning with Non-Negative Risk Estimator (2017, background): “mpirical risk minimizer. Experiments demonstrate that our risk estimator ﬁxes the overﬁtting problem of its unbiased counterparts. 1 Introduction Positive-unlabeled (PU) learning can be dated back to [1,2,3] and has been well studied since then. It mainly focuses on binary classiﬁcation applied to r”
Semi-Supervised Novelty Detection (2010, background): “A theoretical analysis of LPUE is provided by Denis (1998); Denis et al. (2005) from the viewpoint of probably approximately correct (PAC) learnable classes.”
PEBL: Web page classification without negative examples (2004, methodology): “In many real-world learning problems including Web page classification, unlabeled and positive, data are widely available whereas acquiring a reasonable sampling of the negative is impossible or expensive because the negative data set is just the complement of the positive one, and, thus, its probab”
A bagging SVM to learn from positive and unlabeled examples (2010, background): “…et al., 2001; Manevitz and Yousef, 2001; Vert and Vert, 2006; De Bie et al., 2007) Alternatively, the problem of inductive PU learning has been studied on its own from a theoretical viewpoint (Denis et al., 2005; Scott and Blanchard, 2009), and has given rise to a number of speciﬁc algorithms.”
A Survey of Recent Trends in One Class Classification (2009, background): “[27] design an algorithm which is based on positive statistical queries (estimates for probabilities over the set of positive instances) and instance statistical queries (estimates for probabilities over the instance space).”
A survey on semi-supervised learning (2019, background): “We also do not cover learning from positive and unlabelled data, which is a special case of semi-supervised learning where the algorithm has access to a set of unlabelled data points, but where all labelled data points belong to a single class (see, e.g. Liu et al. 2002; Denis et al. 2005; Elkan and”
Learning classifiers from only positive and unlabeled data (2008, background): “Most algorithms based on this assumption need p(y = 1) to be an additional input piece of information; a recent paper emphasizes the importance of p(y = 1) for learning from positive and unlabeled data [4].”