The Value of Unlabeled Data for Classification Problems

By Tong Zhang. International Conference on Machine Learning, 2000

338 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/9755f9993553131e5cc796d34ecaa624fe0ddffa

How researchers see this paper

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

Citations in the corpus

Active Learning Literature Survey (2009, methodology): “Active learning for classiﬁcation tasks has been widely studied (e.g., Cohn et al., 1994; Zhang and Oles, 2000; Guo and Greiner, 2007).”
Adaptive computation and machine learning (1998, methodology): “Generative models are well suited for semi-supervised learning by maximum likelihood, because the likelihood is directly affected by unlabeled data — as opposed to discriminative models, where the associated likelihood is not affected by unlabeled data (Zhang and Oles, 2000).”
Machine learning framework for early MRI-based Alzheimer’s conversion prediction in MCI subjects (2014, unknown): “Generally, unlabeled data improves the classification performance when the assumed model is correct (Zhang and Oles, 2000) and the amount of improvement depends strongly on the number of labeled data and the problem complexity (Cohen et al.”
Interactive deduplication using active learning (2002, background): “Theoretical analysis of this phenomenon can be found in [34].”
Semi-supervised learning by disagreement (2010, unknown): “Later, Zhang and Oles [78] analyzed the value of unlabeled data for parametric models.”
Semi-supervised Learning with Ladder Networks (2015, background): “It is well known (see, e.g., Zhang and Oles, 2000) that when training a probabilistic model that directly estimates P ( y | x ) , unlabeled data cannot help.”
Batch mode active learning and its application to medical image classification (2006, methodology): “In this section, we report our empirical study of batch mode active learning in the application to medical image categorization.”
Improvability Through Semi-Supervised Learning: A Survey of Theoretical Results (2019, background): “Earlier work by Zhang and Oles (2000) distinguishes the same type of models, but the impossibility is about the asymptotic efﬁciency of semi-supervised classiﬁers.”
Learning with Local and Global Consistency (2003, background): “Since labeling often requires expensive human labor, whereas unlabeled data is far easier to obtain, semisupervised learning is very useful in many real-world problems and has recently attracted a considerable amount of research [10].”
Semi-supervised interpolation in an anticausal learning scenario (2015, background): “, k} only (Zhang and Oles, 2000) and X ∈ Rd is a high-dimensional vector.”
Adaptive Active Learning for Image Classification (2013, methodology): “Another class of active learning approaches minimize the generalization error indirectly by reducing model variances, including a statistical approach [4], and a similar approach that selects optimal queries based on Fisher information [35].”
Generalized Expectation Criteria for Semi-Supervised Learning with Weakly Labeled Data (2010, background): “Generalized expectation criteria are unique in that it uses the expected distribution as the sole criterion for optimizing the model parameters on one set of unlabeled data (though it may also use labeled data).”
Building text classifiers using positive and unlabeled examples (2003, background): “Finally, our work is related to learning using a small labeled set and a large unlabeled set [2][3][4][5][10] [11][25][26][28][35].”
Towards Making Unlabeled Data Never Hurt (2011, background): “Many empirical studies, however, show that there are cases in which the use of unlabeled data decreases the performance [7], [11], [13], [14], [16], [20], [36], [47], [50].”