Learning From Noisy Examples

By Dana Angluin, Philip D. Laird. 1988

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

Security analysis of online centroid anomaly detection (2010, background): “On the other hand, several examples of effective attacks have been demonstrated in the context of specific security and spam detection applications (Lowd and Meek, 2005a; Fogla et al., 2006; Fogla and Lee, 2006; Perdisci et al., 2006; Newsome et al., 2006; Nelson et al., 2008), which has motivated…”
Learning to Reweight Examples for Robust Deep Learning (2018, neutral): “Meanwhile, the noisy label problem has been thoroughly studied by the learning theory community (Natarajan et al., 2013; Angluin & Laird, 1988) and practical methods have also been proposed (Reed et al., 2014; Sukhbaatar & Fergus, 2014; Xiao et al., 2015; Azadi et al., 2016; Goldberger &…”
Co-teaching: Robust training of deep neural networks with extremely noisy labels (2018, background): “1 Introduction Learning from noisy labels can date back to three decades ago [1], and still keeps vibrant in recent years [13, 31].”
Pervasive Label Errors in Test Sets Destabilize Machine Learning Benchmarks (2021, background): “Inherent in Q ˜ y,y ∗ is the assumption that noise is class-conditional [1], depending only on the latent true class, not the data.”
Noise-tolerant fair classification (2019, background): “Prominent noise models include symmetric and class-conditional noise (Angluin & Laird, 1988; Bylander, 1994; Blum et al., 1996; Blum & Mitchell, 1998), constant-partition noise (Decatur, 1997; Ralaivola et al., 2006), and instance-dependent noise (Awasthi et al., 2015).”
A high-dimensional Wilks phenomenon (2011, background): “Under the so-called random classification noise (Angluin and Laird 1987; Kearns et al. 1997; Bartlett et al. 2002), noise conditions turn out to be very simple and the sharp estimates on the excess risk derived by Massart and Nedelec (2006) can be injected in Theorems 2 and 5.”
Tri-training: exploiting unlabeled data using three classifiers (2005, methodology): “Inspired by Goldman and Zhou [14], the finding of Angluin and Laird [1] is used in the following analysis.”
Choosing the Sample with Lowest Loss makes SGD Robust (2020, background): “Label noise in deep learning [2, 21, 4] describe diﬀerent techniques to learn in presence of label noise and outliers.”
Confident Learning: Estimating Uncertainty in Dataset Labels (2019, background): “First, Angluin and Laird (1988)’s classification noise process (CNP) provides a starting assumption, that label noise is class-conditional, depending only on the latent true class, not the data.”
A study of the effect of different types of noise on the precision of supervised learning techniques (2010, background): “With respect to other studies of noise in data, in the literature (Sloan 1988; Angluin and Laird 1988; Goldman and Sloan 1995; Sloan 1995; Meeson et al. 1996; Kearns 1998; Nelson 2005), we have observed that many approaches tend to reﬂect and be speciﬁc to the domain and problem.”
Robust data encodings for quantum classifiers (2020, unknown): “[73] D.”
Agnostic Learning of Monomials by Halfspaces Is Hard (2009, methodology): “The Maximum Agreement problem for monotone monomials was shown to be NP-hard by Angluin and Laird [2], and NP-hardness for general monomials was shown by Kearns and Li [28].”
Learning Kernel Perceptrons on Noisy Data Using Random Projections (2007, background): “Definition 2 (CN distributions U [4]).”
Are Anchor Points Really Indispensable in Label-Noise Learning? (2019, background): “Label-noise learning can be dated back to [1] but becomes a more and more important topic recently.”
Ieee Transactions on Pattern Analysis and Machine Intelligence Classification with Noisy Labels by Importance Reweighting (unknown): “|X) can be estimated by a simple probabilistic classification method, where the corresponding link function maps the outputs of the learned predictor to the interval [0] [1] and thus can be interpreted as probabilities.”
Classification in the presence of class noise using a probabilistic Kernel Fisher method (2007, unknown): “The class noise is assumed to have been generated by a classification noise process [21].”
Collaborative recommendation: A robustness analysis (2004, methodology): “In the classification noise model [Angluin and Laird 1988], the noise is modelled with a parameter η as follows: instances are labelled by the target concept C with probability 1−η, and by C with probability η.”
Classification in the Presence of Label Noise: A Survey (2014, background): “In [11], it is shown that the presence of uniform label noise in the probably approximately correct (PAC) framework [117] increases the number of necessary samples for PAC identiﬁ-cation.”