Learning classifiers from only positive and unlabeled data

By Charles Peter Elkan, Keith Noto. 2008

1,201 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Representation & Self-Supervised Learning

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

In Representation & Self-Supervised Learning

Citations in the corpus

D2C: Diffusion-Decoding Models for Few-Shot Conditional Generation (2021, background): “, we merely want to generate new samples similar to given ones), we can train a positive-unlabeled (PU) classifier [33] where we assign “positive” to the new examples and “unlabeled” to training data.”
Seeing through the Human Reporting Bias: Visual Classifiers from Noisy Human-Centric Labels (2015, background): “Bootstrapping [40], semi-supervised learning (SSL) [45, 57] and Positive Unlabeled (PU) learning [10, 28, 30, 34] are other ways of learning from noisy labeled data.”
On the Effectiveness of Linear Models for One-Class Collaborative Filtering (2016, background): “(Elkan and Noto 2008) proved a surprising fact about learning from such data, namely, that for the purposes of ranking , it sufﬁces to simply treat the unlabelled examples as negative, and estimate class-probabilities.”
Differential privacy based on importance weighting (2013, methodology): “A similar idea was used previously by Smith and Elkan (2004) and Elkan and Noto (2008).”
Positive-Unlabeled Learning with Non-Negative Risk Estimator (2017, background): “It mainly focuses on binary classification with application to retrieval and novelty or outlier detection (Elkan & Noto, 2008; Ward et al., 2009; Scott & Blanchard, 2009; Blanchard et al., 2010), while it also has applications in matrix completion (Hsieh et al., 2015) and sequential data (Li et…”
Learning with Confident Examples: Rank Pruning for Robust Classification with Noisy Labels (2017, background): “Typical approaches require averaging predicted probabilities on a holdout set (Liu & Tao, 2016; Elkan & Noto, 2008) tying noise estimation to the accuracy of the predicted probabilities, which in practice may be confounded by added noise or poor model selection.”
Analysis of Learning from Positive and Unlabeled Data (2014, methodology): “In practice, the unknown class prior π may be estimated by the methods proposed in [10, 1, 11].”
Positive-Unlabeled Reward Learning (2019, methodology): “We then show how both issues above can be addressed by reformulating the corresponding (PN) classification problems as PU classification [14, 15, 16].”
Unbiased Recommender Learning from Missing-Not-At-Random Implicit Feedback (2019, background): “R ∈ {0, 1}m×n is a relevance matrix, and each entry Ru,i is a Bernoulli random variable representing the relevance of user u and item i .”
Social Media Text Classification under Negative Covariate Shift (2015, background): “Our work is also related to learning from positive and unlabeled examples, also known as PU learning (Denis, 1998; Yu et al. 2002; Liu et al. 2003; Lee and Liu, 2003; Elkan and Noto, 2008; Li et al. 2010).”
A Review on Multi-Label Learning Algorithms (2014, methodology): “When all the labels are independent, multi-label learning under this case can be decomposed into a series of PU-learning (Positive and Unlabeled data learning) problems [43], where instance without a specific label is considered to be unlabeled.”
deepDR: a network-based deep learning approach to in silico drug repositioning (2019, background): “Motivated by the positive-unlabeled (PU) learning framework, where observed and unobserved entries are penalized differently in the objective (Elkan and Noto, 2008; Hsieh et al., 2014), we introduced a parameter a to balance between positive samples and negative samples.”
Identifying Where to Focus in Reading Comprehension for Neural Question Generation (2017, methodology): “For the related important sentence detection task in text summarization, Yang et al. (2017) therefore propose a two-stage approach (Lee and Liu, 2003; Elkan and Noto, 2008) to augment the set of known summaryworthy sentences.”
Outlier Analysis (2013, methodology): “A number of methods for classification from positive and unlabeled data are discussed in [152], and a good review of the previous work in this area may also be found from the references in this paper.”
Federated Learning with Only Positive Labels (2020, background): “Diﬀerent from PU learning (Liu et al., 2002; Elkan & Noto, 2008; Plessis et al., 2015; Hsieh et al., 2015), in the federated learning setting, the clients do not have access to unlabeled data for both positive and negative classes.”
Partially Supervised Graph Embedding for Positive Unlabelled Feature Selection (2016, background): “Recent years have witnessed a challenging variant of semi-supervised learning, namely positive-unlabelled (PU) learning [Elkan and Noto, 2008].”
Noise-tolerant fair classification (2019, background): “…noise is a special case of MC learning with: πa,corr = (1− ρ+) · πa + ρ− · (1− πa) α = π−1a,corr · (1− πa) · ρ− β = (1− πa,corr)−1 · πa · ρ+ Meanwhile, for PU learning we consider the censoring setting (Elkan & Noto, 2008) which is simply a special case of CCN learning where one of ρ+ and ρ− is 0.”
A bagging SVM to learn from positive and unlabeled examples (2010, methodology): “This approach has been studied, with different weighting schemes, using a logistic regression or a SVM as binary classifier (Liu et al., 2003; Lee and Liu, 2003; Elkan and Noto, 2008).”