Are Anchor Points Really Indispensable in Label-Noise Learning?

By Xiaobo Xia, Tongliang Liu, N. Wang, Bo Han, Chen Gong, Gang Niu, Masashi Sugiyama. Neural Information Processing Systems, 2019

461 citations recorded by the source corpus.

In label-noise learning, \textit{noise transition matrix}, denoting the probabilities that clean labels flip into noisy labels, plays a central role in building \textit{statistically consistent classifiers}. Existing theories have shown that the transition matrix can be learned by exploiting \textit{anchor points} (i.e., data points that belong to a specific class almost surely). However, when there are no anchor points, the transition matrix will be poorly learned, and those current consistent classifiers will significantly degenerate. In this paper, without employing anchor points, we propose a \textit{transition-revision} ($T$-Revision) method to effectively learn transition matrices, leading to better classifiers. Specifically, to learn a transition matrix, we first initialize it by exploiting data points that are similar to anchor points, having high \textit{noisy class posterior probabilities}. Then, we modify the initialized matrix by adding a \textit{slack variable}, which can be learned and validated together with the classifier by using noisy data. Empirical results on benchmark-simulated and real-world label-noise datasets demonstrate that without using exact anchor points, the proposed method is superior to the state-of-the-art label-noise learning methods.

Subfields: Artificial Intelligence, Computational Theory and Mathematics

View Paper: https://www.semanticscholar.org/paper/4f62d8a036f7616510909d4a19db19b735264d9c

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

1. Han-Jia Ye
2. De-chuan Zhan
3. Da-Wei Zhou
4. Qiwen Wang
5. Ziwei Liu
6. Zhi-Hong Qi
7. Liang Liu
8. Haiyang Zhang

In Computational Theory and Mathematics

1. Tongliang Liu
2. Chen Gong
3. Yi Chang
4. De-chuan Zhan
5. Ivor Wai-Hung Tsang
6. Ziwei Liu
7. Han-Jia Ye
8. Gang Niu

Citations in the corpus

• Robust early-learning: Hindering the memorization of noisy labels (2021, methodology): “Some methods among them employ early stopping explicitly or implicitly (Patrini et al., 2017; Xia et al., 2019).”
• Early-Learning Regularization Prevents Memorization of Noisy Labels (2020, methodology): “methods explicitly correct the loss function to take into account the noise distribution, represented by a transition matrix of mislabeling probabilities [31, 12, 46, 39].”
• From Synthetic to Real: Unsupervised Domain Adaptation for Animal Pose Estimation (2021, background): “To reduce the negative effect of noisy labels, some approaches focus on training noise robust models by designing robust losses [12, 34, 44] or by correcting the loss with a transition matrix [27, 13, 37].”
• DualGraph: A graph-based method for reasoning about label noise (2021, methodology): “The early methods focus on estimating the label transition matrix [18, 19, 21, 31].”
• Learning with Noisy Labels Revisited: A Study Using Real-World Human Annotations (2021, methodology): “…of approaches modify the loss of image samples by multiplying an estimated noise transition matrix [Patrini et al., 2017, Hendrycks et al., 2018, Xia et al., 2019, Yao et al., 2020, Zhu et al., 2021c], re-weight the loss to encourage deep neural nets to fit on correct labels [Liu and Tao,…”
• Deep Class-Incremental Learning: A Survey (2023, background): “, few-shot [204], imbalanced [136], with noisy label [205], weak supervised [206], multi modal [207], concept drift [13], novel classes [208], [209] etc.”
• The Emerging Trends of Multi-Label Learning (2020, background): “Since learning from label noise [162] have been a hot topic”
• Image Classification with Deep Learning in the Presence of Noisy Labels: A Survey (2019, methodology): “Iterative calculation: EM [48, 49, 27], fully connected layer [50], anchor point estimate [51] Drichlet-distribution [52] c.”
• Beyond Class-Conditional Assumption: A Primary Attempt to Combat Instance-Dependent Label Noise (2020, unknown): “Since the noise transition process is fully specified by a matrix M , one can mitigate the effect of CCN by modeling M (Patrini et al. 2017; Hendrycks et al. 2018; Han et al. 2018a; Xia et al. 2019; Yao et al. 2019).”
• Combating Noisy Labels by Agreement: A Joint Training Method with Co-Regularization (2020, methodology): “The early methods focus on estimating the label transition matrix [24, 25, 28, 37].”