Unlabeled data: Now it helps, now it doesn't

By Aarti Singh, Robert D. Nowak, Xiaojin Zhu. Neural Information Processing Systems, 2008

310 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/5a002b24c0a8a63527a8c0e0681d4747e227aa6f

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

Understanding Self-Training for Gradual Domain Adaptation (2020, background): “Other theory in semi-supervised learning (Rigollet, 2007; Singh et al., 2008; Ben-David et al., 2008) does not analyze domain shift.”
A Semi-Supervised Learning Approach to Differential Privacy (2013, background): “Figure 10 shows the decrease in error rate as the number of labeled instances increases in the Mushroom data set.”
A Survey on Deep Semi-Supervised Learning (2021, background): “Although it is remarkable that unlabeled data can improve learning performance under appropriate assumptions or conditions, some empirical studies [132], [133], [134] have shown that the use of unlabeled datamay lead to performance degeneration,making the generalization performance even worse than a”
Generalizing from a Few Examples (2019, methodology): “or reduce S. Based on how prior knowledge is used, we categorize these works into kinds: •Data: methods that use prior knowledge to augment Dtrain of I samples to I˜samples. In this way, S can be met [12, 107]. And one can obtain a more accurate empirical risk minimizerh I˜ as shown in Figure 2(a). ”
Projected estimators for robust semi-supervised classification (2016, methodology): “The goal of semi-supervised learning is to improve the classiﬁcation decision function f : R d → R using the unlabeled information in X u as compared to the case where we do not have these unlabeled objects.”
Causality matters in medical imaging (2019, background): “Under certain conditions, unlabelled data can be shown to have a regularizing effect, potentially boosting the accuracy of an imperfect model by lowering its variance 13 , and may reduce the amount of labelled data required to achieve a given performance level 14,15 .”
A survey on semi-supervised learning (2019, background): “As is the case for most semi-supervised learning methods, S3VMs are not guaranteed to perform better than their supervised counterparts (Singh et al. 2009).”
Survey on distance metric learning and dimensionality reduction in data mining (2014, background): “There are also research on when the unlabeled data could be helpful (Singh et al. 2008).”
Contrastive Pessimistic Likelihood Estimation for Semi-Supervised Classification (2015, methodology): “We show that semi-supervised estimates can be constructed that are essentially closer to the estimates that would be obtained when also all the labels for all unlabeled data would be available in the training phase.”
Introduction to Semi-Supervised Learning (2009, background): “• advances in learning theory for semi-supervised learning [4, 9, 46, 63, 143, 161, 162, 164].”
Dash: Semi-Supervised Learning with Dynamic Thresholding (2021, background): “…improve the learning performance [Berthelot et al., 2019b,a, Sohn et al., 2020], however, a series of theoretical studies [Ben-David et al., 2008, Singh et al., 2009, Li and Zhou, 2011b, Balcan and Blum, 2005] have demonstrated that this success is highly relying on a necessary condition that…”
Improvability Through Semi-Supervised Learning: A Survey of Theoretical Results (2019, background): “Singh et al. (2008) propose a different formalization of the cluster assumption, one that allows to distinguish cases where SSL does help and where not.”