On Symmetric Losses for Learning from Corrupted Labels

By Nontawat Charoenphakdee, Jongyeong Lee, Masashi Sugiyama. 2019

116 citations recorded by the source corpus.

This paper aims to provide a better understanding of a symmetric loss. First, we emphasize that using a symmetric loss is advantageous in the balanced error rate (BER) minimization and area under the receiver operating characteristic curve (AUC) maximization from corrupted labels. Second, we prove general theoretical properties of symmetric losses, including a classification-calibration condition, excess risk bound, conditional risk minimizer, and AUC-consistency condition. Third, since all nonnegative symmetric losses are non-convex, we propose a convex barrier hinge loss that benefits significantly from the symmetric condition, although it is not symmetric everywhere. Finally, we conduct experiments to validate the relevance of the symmetric condition.

Subfields: Artificial Intelligence

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. Xingjun Ma
2. Yisen Wang
3. James Bailey
4. Xia Hu
5. Qingquan Song
6. Yi-Wei Chen
7. Görkem Algan
8. Ilkay Ulusoy

Citations in the corpus

• On Robustness of Neural Architecture Search Under Label Noise (2020, background): “For other symmetric loss functions (Charoenphakdee et al., 2019), we leave them for future work.”
• Normalized Loss Functions for Deep Learning with Noisy Labels (2020, background): “We refer the reader to (Charoenphakdee et al., 2019) for more discussions of other theoretical properties such as classification calibration.”
• AUC Maximization in the Era of Big Data and AI: A Survey (2022, unknown): “As a result, AUC maximization can be formulated as min Barrier Hinge ℓ ( 𝑠 ) = max (− 𝑏 ( 𝑟 + 𝑠 ) + 𝑟, max ( 𝑏 ( 𝑠 − 𝑟 ) ,𝑟 − 𝑠 )) 𝑟 > 0 ,𝑏 > 0 Noisy Labels [22] Sigmoid )) and (one-way) pAUC maximization can be formulated as min and two-way pAUC maximization can be formulated as min Reg”
• Beyond Synthetic Noise: Deep Learning on Controlled Noisy Labels (2019, background): “…label is also commonly used as a benchmark to evaluate robust learning methods in a noise-control setting e.g. in (Vahdat, 2017; Shu et al., 2019; Jiang et al., 2018; Ma et al., 2018; Han et al., 2018b; Van Rooyen et al., 2015; Li et al., 2019; Arazo et al., 2019; Charoenphakdee et al., 2019).”
• Image Classification with Deep Learning in the Presence of Noisy Labels: A Survey (2019, background): “Robust Losses Non-convex loss functions [111, 112, 113], 0-1 loss surrogate [114], MAE [115], IMEA [116] Generalized cross-entropy [117], symmetric loss [118], unbiased estimator [119], modified cross-entropy for omission [120], information theoric loss [121] linear-odd losses [122], classification ”
• Noise-tolerant fair classification (2019, background): “, 2015), which are prominent noise models that have seen sustained study in recent years (Jain et al., 2017; Kiryo et al., 2017; van Rooyen & Williamson, 2018; Katz-Samuels et al., 2019; Charoenphakdee et al., 2019).”
• Does label smoothing mitigate label noise? (2020, methodology): “Indeed, label smoothing is conspicuously absent in most treatments of the noisy label problem [Patrini et al., 2016, Han et al., 2018b, Charoenphakdee et al., 2019, Thulasidasan et al., 2019, Amid et al., 2019].”