Ieee Transactions on Pattern Analysis and Machine Intelligence Classification with Noisy Labels by Importance Reweighting

By Tongliang Liu, Dacheng Tao

Subfields: Natural Language Processing, Statistical Learning Theory

How researchers see this paper

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

In Natural Language Processing

1. Dacheng Tao
2. Gang Niu
3. Bo Han
4. Changxing Ding
5. Mingming Gong
6. Tongliang Liu
7. Jessa Bekker
8. Jesse Davis

In Statistical Learning Theory

1. Dacheng Tao
2. Tongliang Liu
3. Gang Niu
4. Bo Han
5. Changxing Ding
6. Mingming Gong
7. Jessa Bekker
8. Jesse Davis

Citations in the corpus

• Unsupervised Embedding Learning via Invariant and Spreading Instance Feature (2019, background): “For negative separation: since unlabelled data are usually highly imbalanced [27, 49], the number of negative samples for each image instance is much larger than that of positive samples.”
• Co-teaching: Robust training of deep neural networks with extremely noisy labels (2018, methodology): “If is not known in advanced, can be inferred using validation sets [23, 43].”
• Generalized Cross Entropy Loss for Training Deep Neural Networks with Noisy Labels (2018, background): “[23] derived an optimal importance weighting scheme for noise-robust classification.”
• Learning from Noisy Labels with Distillation (2017, methodology): “proposed an importance re-weighting method to deal with random classification noise [12], where the level of label noise is estimated from a pretrained classifier.”
• Learning with Noisy Labels Revisited: A Study Using Real-World Human Annotations (2021, methodology): “, 2021c], re-weight the loss to encourage deep neural nets to fit on correct labels [Liu and Tao, 2015], propose robust loss functions [Natarajan et al.”
• Making Deep Neural Networks Robust to Label Noise: A Loss Correction Approach (2016, methodology): “Yet, inter-estingly, recent work has provided practical algorithms for estimating the noise rates [36, 34, 21, 26, 31]; remarkably, this is achievable with absolutely no knowledge of ground truth labels .”
• Are Anchor Points Really Indispensable in Label-Noise Learning? (2019, background): “Œus, anchor points are assumed to exist but unknown in datasets, which can be identi€ed either theoretically [Liu and Tao, 2016] or heuristically [Patrini et al.”
• Feature Selection Based on Structured Sparsity: A Comprehensive Study (2017, methodology): “SVM is a widely used classification technique [128].”
• A Novel Self-Supervised Re-labeling Approach for Training with Noisy Labels (2019, background): “Noise rate estimation is also addressed in the works of [23] [18] and [33].”
• Noise-tolerant fair classification (2019, background): “Algorithms that deal with CCN corruption have been well studied (Natarajan et al., 2013; Liu & Tao, 2016; Northcutt et al., 2017).”
• Cleaning Crowdsourced Labels Using Oracles For Statistical Classification (2018, methodology): “[25] use importance reweighting to ensure that any surrogate loss function can be used for classification, proving that the label noise involved in training will not affect the search for an optimal classifier.”
• Simple and Effective Regularization Methods for Training on Noisily Labeled Data with Generalization Guarantee (2019, background): “A large body of work proposed various methods for training with mislabeled examples, such as estimating noise distribution (Liu and Tao, 2015) or confusion matrix (Sukhbaatar et al.”
• Deep learning with noisy labels: exploring techniques and remedies in medical image analysis (2019, background): “Furthermore, it has been shown that proper re-weighting of training samples can improve the robustness of many loss functions to label noise [44],”
• Robust Active Label Correction (2018, methodology): “…model is well-studied 1For all subsequent considerations it is actually not necessary to be able to obtain the true label; a significantly more accurate label is sufficient. in the label noise literature (Bootkrajang and Kabán, 2012; Natarajan et al., 2013; Menon et al., 2015; Liu and Tao, 2016).”
• Self-Supervised Representation Learning by Rotation Feature Decoupling (2019, methodology): “Then the mislabeled rate can be used to handle noisy observed negative examples by various ways such as excluding examples with low confidence [40], labeling examples with high confidence [49, 23, 19] or reweighting examples [14, 35, 32].”
• How does Disagreement Help Generalization against Label Corruption? (2019, methodology): “In the noise rate estimation category, both [23] and [16] proposed a class-probability estimator using order statistics on the range of scores.”