Adversarial Robustness under Long-Tailed Distribution

Wu, Tong; Liu, Ziwei; Huang, Qingqiu; Wang, Yu; Lin, Dahua

doi:10.1109/cvpr46437.2021.00855

Cited by 55 publications

(22 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In addition, there are several approaches that employ Causal Inference [145], Adversarial Training [171], Distributional Robust Optimization (DRO) [40] etc. to solve the long-tailed problem.…”

Section: More Methodsmentioning

confidence: 99%

“…Wu et al [171] find that long-tailed data have a negative impact on adversarial robustness and that the natural accuracy loss of the tail classes is further magnified in adversarial training. Meanwhile, they argue that suitable features as well as classifier embedding help to reduce the boundary error, and the combination of long-tailed recognition methods with the adversarial training framework helps to improve the natural accuracy.…”

Section: More Methodsmentioning

confidence: 99%

See 1 more Smart Citation

A Survey on Long-Tailed Visual Recognition

et al. 2022

View full text Add to dashboard Cite

The heavy reliance on data is one of the major reasons that currently limit the development of deep learning. Data quality directly dominates the effect of deep learning models, and the long-tailed distribution is one of the factors affecting data quality. The long-tailed phenomenon is prevalent due to the prevalence of power law in nature. In this case, the performance of deep learning models is often dominated by the head classes while the learning of the tail classes is severely underdeveloped. In order to learn adequately for all classes, many researchers have studied and preliminarily addressed the long-tailed problem. In this survey, we focus on the problems caused by long-tailed data distribution, sort out the representative long-tailed visual recognition datasets and summarize some mainstream long-tailed studies. Specifically, we summarize these studies into ten categories from the perspective of representation learning, and outline the highlights and limitations of each category. Besides, we have studied four quantitative metrics for evaluating the imbalance, and suggest using the Gini coefficient to evaluate the long-tailedness of a dataset. Based on the Gini coefficient, we quantitatively study 20 widely-used and large-scale visual datasets proposed in the last decade, and find that the long-tailed phenomenon is widespread and has not been fully studied. Finally, we provide several future directions for the development of long-tailed learning to provide more ideas for readers.

show abstract

Section: More Methodsmentioning

confidence: 99%

Section: More Methodsmentioning

confidence: 99%

A Survey on Long-Tailed Visual Recognition

et al. 2022

View full text Add to dashboard Cite

show abstract

“…We also investigate the combination of the state-of-the-art centralized AT methods with FL, i.e., we apply standard PGD [21], TRADES [37], MART [31]), and GAIRAT [38] to FL, and term them as FedPGD, FedTRADES, FedMART, and FedGAIRAT. Additionally, we apply the state-of-the-art long-tail learning methods (LogitAdj [23] and RoBal [33]) to FAT, and term them as FedLogitAdj and FedRoBal.…”

Section: Details Of Calfatmentioning

confidence: 99%

“…)HG*$,5$7 )HG5%1 &DO)$7RXUV Comparison with state-of-the-art long-tail learning methods We also adapt the losses of long-tail learning methods (LogitAdj [23] and RoBal [33]) to FAT (namely FedLogitAdj and FedRoBal) for both the local inner and outer optimization. As shown in Table 3, both methods have lower natural and robust accuracy than our CalFAT.…”

Section: Andrppxqlfdwlrqurxqg 5rexvwdffxudf\mentioning

confidence: 99%

CalFAT: Calibrated Federated Adversarial Training with Label Skewness

Chen¹,

Liu²,

Ma³

et al. 2022

Preprint

View full text Add to dashboard Cite

Recent studies have shown that, like traditional machine learning, federated learning (FL) is also vulnerable to adversarial attacks. To improve the adversarial robustness of FL, few federated adversarial training (FAT) methods have been proposed to apply adversarial training locally before global aggregation. Although these methods demonstrate promising results on independent identically distributed (IID) data, they suffer from training instability issues on non-IID data with label skewness, resulting in much degraded natural accuracy. This tends to hinder the application of FAT in real-world applications where the label distribution across the clients is often skewed. In this paper, we study the problem of FAT under label skewness, and firstly reveal one root cause of the training instability and natural accuracy degradation issues: skewed labels lead to non-identical class probabilities and heterogeneous local models. We then propose a Calibrated FAT (CalFAT) approach to tackle the instability issue by calibrating the logits adaptively to balance the classes. We show both theoretically and empirically that the optimization of CalFAT leads to homogeneous local models across the clients and much improved convergence rate and final performance.

show abstract

“…Due to the paucity of training examples, generalisation for tail classes is challenging; moreover, naïve learning on such data is susceptible to an undesirable bias towards head classes. Recently, long-tailed learning (LTL) has gained renewed interest in the context of deep neural networks [6,7,8,9,10,11,12,13]. Two active strands of work involve normalisation of the classifier's weights, and modification of the underlying loss to account for different class penalties.…”

Section: Introductionmentioning

confidence: 99%

Robust Long-Tailed Learning under Label Noise

Wei¹,

Shi²,

Tu³

et al. 2021

Preprint

View full text Add to dashboard Cite

Long-tailed learning has attracted much attention recently, with the goal of improving generalisation for tail classes. Most existing works use supervised learning without considering the prevailing noise in the training dataset. To move long-tailed learning towards more realistic scenarios, this work investigates the label noise problem under long-tailed label distribution. We first observe the negative impact of noisy labels on the performance of existing methods, revealing the intrinsic challenges of this problem. As the most commonly used approach to cope with noisy labels in previous literature, we then find that the small-loss trick fails under long-tailed label distribution. The reason is that deep neural networks cannot distinguish correctly-labeled and mislabeled examples on tail classes. To overcome this limitation, we establish a new prototypical noise detection method by designing a distance-based metric that is resistant to label noise. Based on the above findings, we propose a robust framework, ROLT, that realizes noise detection for long-tailed learning, followed by soft pseudo-labeling via both label smoothing and diverse label guessing. Moreover, our framework can naturally leverage semi-supervised learning algorithms to further improve the generalisation. Extensive experiments on benchmark and real-world datasets demonstrate the superiority of our methods over existing baselines. In particular, our method outperforms DivideMix by 3% in test accuracy. Source code will be released soon.

show abstract

Adversarial Robustness under Long-Tailed Distribution

Cited by 55 publications

References 18 publications

A Survey on Long-Tailed Visual Recognition

A Survey on Long-Tailed Visual Recognition

CalFAT: Calibrated Federated Adversarial Training with Label Skewness

Robust Long-Tailed Learning under Label Noise

Contact Info

Product

Resources

About