Long-Tailed Classification with Diversified Classifier Under Stability = 안정성이 보장된 다양화된 롱테일 분류기 설계

다국어 초록 (Multilingual Abstract)

This dissertation explores how to train long-tailed classifiers with diversity and stability to prevent inadvertent degradation of performance in majority classes to enhance minority class performance. Deriving how the degradation of performance occurs from balanced loss and an ensemble of multiple experts that researched so far, we propose two approaches to solve the problem.

First, we introduce a novel regularization technique named Logit Weight Repulsion (LWR). LWR aims to push apart the logit weight vectors associated with different classes, thereby expanding the separation between these vectors. We assessed the effectiveness of LWR on standard datasets, demonstrating that our approach competes well with current state-of-the-art methods in long-tailed classification tasks. Importantly, LWR enhances performance in minority classes with less compromisation of the performance of majority classes.

Second, we propose Consensus and Excessive Diversity Aversion Learning (CEDA). Recent expert ensemble methods for long-tailed recognition encourage diversity by maximizing KL divergence between the predictions of experts. However, the excessive diversity using KL divergence, which has no upper bound, induces inaccurate predictions from experts. To address this issue, we propose a new learning method for expert ensemble, which obtains the consensus by aggregating the predictions of experts (Consensus) and maximizes the expected prediction accuracy of each expert without excessive diversity from the consensus (Excessive Diversity Aversion). To implement this learning scheme, we propose a new loss derived from Rényi Divergence. We provide both empirical and theoretical analysis of the proposed method along with a stability guarantee, which is not guaranteed in the existing methods. Thanks to this stability, the proposed method continues to improve performance as the number of experts increases, while the existing methods do not. The proposed method achieves state-of-the-art performance for any number of experts. Furthermore, the proposed method operates robustly even when evaluated by varying the imbalance factor.

Finally, as a future work, we research on how to get maximized consensus from multiple experts. Existing methods to ensemble multiple experts show degradation in performance or involve test data during the training. Therefore, we will design the consensus function that receives predictions from all experts and outputs results that maximize performance.

국문 초록 (Abstract)

이 논문에서는 소수 클래스 성능을 향상시키기 위해 다수 클래스의 의도하지 않은 성능 저하를 방지하기 위해 다양성과 안정성을 갖춘 롱테일 분류기를 학습하는 방법을 탐구한다. 지금까지 연구한 다수의 전문가들의 균형손실함수와 앙상블이 어떻게 의도하지 않은 성능 저하를 일으키는지 도출하고, 문제 해결을 위한 두 가지 접근 방식을 제안한다.

먼저, 우리는 한 클래스의 로짓 가중치 벡터가 다른 클래스의 벡터를 밀어내도록 장려하는 LWR(로짓 가중치 반발)이라는 새로운 정규화 항을 제안한다. 이러한 반발력은 각 클래스의 로짓 벡터 사이의 영역을 확대한다. 제안된 LWR 정규화 도구는 벤치마크 데이터 세트에서 평가되었으며, 우리의 방법은 롱테일 분류에 대한 최첨단 성능과 경쟁적인 성능을 보여준다. 특히, LWR은 다수 클래스의 성능을 희생하지 않고 오히려 증가시키는 동시에 소수 클래스의 성능 향상을 달성한다.

둘째, 우리는 CEDA(합의 및 과도한 다양성 회피 학습)을 제안한다. CEDA은 베팅 이론에서 영감을 받아 본 논문에서 제안한 과도한 다양성 회피 손실을 사용해 학습하는 전략을 사용한다. 경제학과 여러 전문가 양성 간의 격차를 해소하면서 CEDA에 대한 실증적, 이론적 분석과 안정성 보장을 모두 제공한다. CEDA은 긴 꼬리 분류 벤치마크에서 최첨단 방법보다 성능이 뛰어나다. 게다가 안정성 덕분에 전문가를 많이 사용할수록 성능도 좋아진다.

마지막으로 향후 연구로는 여러 전문가의 합의를 극대화할 수 있는 방법을 연구한다. 여러 전문가를 앙상블하는 기존 방법은 성능 저하를 나타내거나 훈련 중에 테스트 데이터를 포함한다. 따라서 모든 전문가의 예측을 받아 성능을 극대화하는 결과를 출력하는 합의 기능을 설계하고자 한다.

목차 (Table of Contents)

• Abstract i
• Contents iii
• List of Tables vi
• List of Figures xi
• 1 Introduction 1
• 1.1 Motivation and Background 1
• 1.2 Problem Statements 2
• 1.3 Contributions 3
• 2 Related Works 5
• 2.1 Balanced Loss 6
• 2.2 Re-sampling 8
• 2.3 Ensemble of Multiple Experts 9
• 2.4 Other Methods For Long-Tailed Classification 11
• 2.5 Regularizer in Long-Tailed Object Classification 13
• 2.6 Information Theory in Economics 14
• 3 Logit Weight Repulsion Regularizer 16
• 3.1 Overview 16
• iii
• 3.2 Analysis of Fatal Effect of Cross Entropy Loss in Long-Tailed Classi-
• fication 19
• 3.3 Analysis of Balanced Loss 19
• 3.4 Logit Weight Repulsion (LWR) 25
• 3.5 Training Schemes 28
• 3.5.1 Implementation Tricks 28
• 3.5.2 Adaptive Weight for Regularization 28
• 3.6 Experiments 29
• 3.6.1 Dataset 29
• 3.6.2 Implementation Details 31
• 3.6.3 Evaluation Metrics 32
• 3.6.4 Benchmark results 33
• 3.6.5 Ablation Studies 40
• 3.7 Limitation 47
• 3.8 Summary 47
• 3.9 Future Works 48
• 3.9.1 Designing adaptive ηi,j for LWR 48
• 3.9.2 Application to Long-Tailed Object Detection 49
• 4 Consensus and Excessive Diversity Aversion 50
• 4.1 Overview 50
• 4.2 Preleminaries 50
• 4.2.1 Rényi Divergence 50
• 4.2.2 Divergence as a Measure of Expected Return 51
• 4.3 Proposed Method 52
• 4.3.1 Consensus and Excessive Diversity Aversion 52
• 4.3.2 Behavior of Excessive Diversity Aversion Loss 55
• 4.3.3 Stability Guarantee of EDA Loss 57
• 4.3.4 Individualized Training of Each Experts 59
• iv
• 4.3.5 Class Prior Adjustment 60
• 4.4 Expertiments and Results 60
• 4.4.1 Implementation Details 60
• 4.4.2 ImageNet-LT 62
• 4.4.3 CIFAR-100 LT 63
• 4.4.4 iNaturalist2018 63
• 4.4.5 Evaluation Metric 64
• 4.4.6 Comparison with State-of-the-arts 68
• 4.4.7 Ablation Studies 76
• 4.5 Limitation 81
• 4.6 Summary 81
• 4.7 Future Works 82
• 4.7.1 Designing adaptive ρ for CEDA 82
• 4.7.2 Design of Advanced Consensus Function for Training Phase 82
• 4.7.3 Using Other Aggregation For Inference Phase 83
• 5 Conclusion 84
• Abstract (In Korean) 103

참고문헌 (Reference)

1. Risk aversion, Jan Werner, The new Palgrave dictionary of economics, , 2008

2. Deep mutual learning, Ying Zhang, TimothyMHospedales and, Huchuan Lu, Tao Xiang, In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4320–4328, , 2018

3. Multi-class adaboost, Saharon Rosset, Hui Zou, Ji Zhu and, Trevor Hastie, 2(3):349–360, , 2009

4. Ensemble learningA survey, Omer Sagi and, Lior Rokach, Wiley interdisciplinary reviews: data mining and knowledge discovery, 8(4):e1249, , 2018

5. Learning to model the tail, Martial Hebert, Yu-Xiong Wang, Deva Ramanan and, Advances in Neural Information Processing Systems, , 2017

6. Mathematics for economists, Carl P Simon, Lawrence Blume et al, volume 7 Norton New York, , 1994

7. Cours d´economie politique, Vilfredo Pareto, volume 1 Librairie Droz, , 1964

8. Early stopping-but when? In, Lutz Prechelt, Neural Networks: Tricks of the trade, pages 55–69. Springer, , 2002

9. On information and sufficiency, Richard A Leibler, Solomon Kullback and, 22(1):79–86, , 1951

10. Parametric contrastive learning, Shu Liu, Zhisheng Zhong, Jiaya Jia, Bei Yu and, Jiequan Cui, In Proceedings of the IEEE/CVF international conference on computer vision, pages 715–724, , 2021

11. Elasticity theory of structuring, Andrei N Soklakov, arXiv preprint arXiv:1304.7535, , 2013

12. Deep long-tailed learningA survey, Jiashi Feng, Bryan Hooi, Shuicheng Yan and, Yifan Zhang, Bingyi Kang, 2023, , 2023

13. A tutorial on bayesian optimization, Peter I Frazier, arXiv preprint arXiv:1807.02811, , 2018

14. Aspects of the theory of risk-bearing, Kenneth Joseph Arrow et al, No Title, , 1965

15. Focal loss for dense object detection, Kaiming He and, Piotr Doll´ar, Ross Girshick, Priya Goyal, Tsung-Yi Lin, In Proceedings of the IEEE international conference on computer vision, pages 2980–2988, , 2017

16. On measures of entropy and information, Alfr´ed R´enyi, In Proceedings of the Fourth Berkeley Symposium on Mathematical Statistics and Probability, Volume 1: Contributions to the Theory of Statistics, volume 4, pages 547–562. University of California Press, , 1961

17. Trustworthy long-tailed classification, Bolian Li, Zongbo Han, Haining Li, Changqing Zhang, Huazhu Fu and, In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6970–6979, , 2022

18. Long-tail learning via logit adjustment, Sadeep Jayasumana, Sanjiv Kumar, Aditya Krishna Menon, Ankit Singh Rawat, Himanshu Jain, Andreas Veit and, arXiv preprint arXiv:2007.07314, , 2020

19. mixupBeyond empirical risk minimization, Hongyi Zhang, Yann N Dauphin and, Moustapha Cisse, David Lopez-Paz, arXiv preprint arXiv:1710.09412, , 2017

20. A new interpretation of information rate, John L Kelly, 35(4):917–926, , 1956

21. Deep learning face attributes in the wild, Xiaoou Tang, Ziwei Liu, Xiaogang Wang and, Ping Luo, In Proceedings of the IEEE international conference on computer vision, pages 3730–3738, , 2015

22. A new metric for probability distributions, Dominik Maria Endres and, Johannes E Schindelin, 49(7):1858–1860, , 2003

23. Longtailed recognition via weight balancing, Deva Ramanan and, Shu Kong, Shaden Alshammari, Yu-Xiong Wang, In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, , 2022

24. Mutual learning for long-tailed recognition, Eunji Jun, Changhwa Park, Junho Yim and, In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 2675–2684, 2023, , 2023

25. Deep residual learning for image recognition, Jian Sun, Shaoqing Ren and, Kaiming He, Xiangyu Zhang, In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, , 2016

26. Distilling the knowledge in a neural network, Geoffrey Hinton, Jeff Dean, Oriol Vinyals and, arXiv preprint arXiv:1503.02531, , 2015

27. Generative adversarial minority oversampling, Shounak Datta and, Sankha Subhra Mullick, Swagatam Das, In Proceedings of the IEEE/CVF international conference on computer vision, pages 1695–1704, , 2019

28. Risk aversion in the small and in the large In, John W Pratt, Uncertainty in economics, pages 59–79. Elsevier, , 1978

29. Alpha-beta divergence for variational inference, Jean-Baptiste Regli and, Ricardo Silva, arXiv preprint arXiv:1805.01045, , 2018

30. Feature space augmentation for long-tailed data, Shaopeng Liu and, Haibin Ling, Xiao Bian, Peng Chu, In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, Proceedings, Part XXIX 16, pages 694–710. Springer, , 2020

31. Smotesynthetic minority over-sampling technique, Kevin W Bowyer, Nitesh V Chawla, W Philip Kegelmeyer, Lawrence O Hall and, 16:321–357, , 2002

32. A simple weight decay can improve generalization, Anders Krogh and, John Hertz, Advances in neural information processing systems, 4, , 1991

33. A survey on contrastive self-supervised learning, Fillia Makedon, Ashish Jaiswal, Debapriya Banerjee and, Ashwin Ramesh Babu, Mohammad Zaki Zadeh, 9(1):2, , 2020

34. Divergence measures based on the shannon entropy, Jianhua Lin, 37(1):145–151, , 1991

35. Posterior re-calibration for imbalanced datasets, Yen-Chang Hsu and, Zsolt Kira, Junjiao Tian, Yen-Cheng Liu, Nathaniel Glaser, Advances in Neural Information Processing Systems, 33:8101–8113, , 2020

36. Imagenet large scale visual recognition challenge, Aditya Khosla, Zhiheng Huang, Sean Ma, Sanjeev Satheesh, Jia Deng, Hao Su, Olga Russakovsky, Jonathan Krause, Andrej Karpathy, Michael Bernstein et al, 115:211–252, , 2015

37. Improving calibration for long-tailed recognition, Shu Liu and, Zhisheng Zhong, Jiequan Cui, Jiaya Jia, Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 16489–16498, , 2021

38. Seesaw loss for long-tailed instance segmentation, Chen Change Loy and, Yuhang Cao, Yuhang Zang, Jiaqi Wang, Wenwei Zhang, Tao Gong, Kai Chen, Dahua Lin, Ziwei Liu, Jiangmiao Pang, In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9695–9704, , 2021

39. ResltResidual learning for long-tailed recognition, Jiequan Cui, Zhuotao Tian, Shu Liu, Jiaya Jia, Zhisheng Zhong and, IEEE Transactions on Pattern Analysis and Machine Intelligence, , 2022

40. R´enyi divergence and kullback-leibler divergence, Tim Van Erven and, Peter Harremos, 60(7):3797–3820, , 2014

41. AutoaugmentLearning augmentation policies from data, Quoc V Le, Ekin D Cubuk, Vijay Vasudevan and, Dandelion Mane, Barret Zoph, arXiv preprint arXiv:1805.09501, , 2018

42. Equalization loss for long-tailed object recognition, Jingru Tan, Quanquan Li, Wanli Ouyang, ChangbaoWang, Junjie Yan, Buyu Li, Changqing Yin and, In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11662–11671, , 2020

43. Large-scale long-tailed recognition in an open world, Ziwei Liu, Boqing Gong and, Xiaohang Zhan, Jiayun Wang, Zhongqi Miao, Stella X Yu, In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2537–2546, , 2019

44. Adversarial robustness under long-tailed distribution, Yu Wang and, Ziwei Liu, Qingqiu Huang, Dahua Lin, Tong Wu, In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8659–8668, , 2021

45. Distributional robustness loss for long-tail learning, Gal Chechik, Dvir Samuel and, Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 9495–9504, , 2021

46. Exploring the limits of weakly supervised pretraining, Ashwin Bharambe and, Manohar Paluri, Vignesh Ramanathan, Kaiming He, Dhruv Mahajan, Laurens van der Maaten, Yixuan Li, Ross Girshick, In Proceedings of the European conference on computer vision (ECCV), pages 181–196, , 2018

47. Learning fast sample re-weighting without reward data, Zizhao Zhang and, Tomas Pfister, Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 725–734, , 2021

48. Exploratory undersampling for class-imbalance learning, Jianxin Wu and, Zhi-Hua Zhou, Xu-Ying Liu, IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), 39(2):539–550, , 2008

49. Balanced metasoftmax for long-tailed visual recognition, Shuai Yi et al, Xiao Ma, Cunjun Yu, Jiawei Ren, Haiyu Zhao, 33:4175–4186, , 2020

50. Classbalanced loss based on effective number of samples, Menglin Jia, Serge Belongie, Yang Song and, Tsung-Yi Lin, Yin Cui, In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9268– 9277, , 2019

51. Adaboost. rta boosting algorithm for regression problems, Durga L Shrestha, Dimitri P Solomatine and, In 2004 IEEE international joint conference on neural networks (IEEE Cat. No. 04CH37541), volume 2, pages 1163–1168. IEEE, , 2004

52. LvisA dataset for large vocabulary instance segmentation, Agrim Gupta, Ross Girshick, Piotr Dollar and, In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5356–5364, , 2019

53. Mesaboost ensemble imbalanced learning with meta-sampler, Yi Chang, Jing Jiang, Jiang Bian and, Wei Cao, PengfeiWei, Zhining Liu, Advances in neural information processing systems, 33:14463–14474, , 2020

54. The pascal visual object classes challengeA retrospective, Luc Van Gool, John Winn and, SM Ali Eslami, Christopher KI Williams, Andrew Zisserman, Mark Everingham, 111:98–136, , 2015

55. Experimental perspectives on learning from imbalanced data, Amri Napolitano, Taghi M Khoshgoftaar and, Jason Van Hulse, In Proceedings of the 24th international conference on Machine learning, pages 935–942, , 2007

56. Learning deep representation for imbalanced classification, C. C. Loy and, X. Tang, Y. Li, C. Huang, In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), , 2016

57. The robust minimum cost consensus model with risk aversion, Ying Ji, Ripeng Huang, Huanhuan Li and, Shaojian Qu, Huijie Zhang, 587:283–299, , 2022

58. Influencebalanced loss for imbalanced visual classification, Jongin Lim, Seulki Park, Younghan Jeon and, Jin Young Choi, In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), , 2021

59. Local and global logit adjustments for long-tailed learning, Daniel Du and, XuWang, Min Zheng, Jingna Sun, Li Chen, Yingfan Tao, Hao Yang, Wenming Yang, In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 11783–11792, 2023, , 2023

60. M2mImbalanced classification via major-to-minor translation, Jaehyung Kim, Jinwoo Shin, Jongheon Jeong and, In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 13896–13905, , 2020

61. Aggregated residual transformations for deep neural networks, Ross Girshick, Zhuowen Tu and, Piotr Doll´ar, Kaiming He, Saining Xie, In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1492– 1500, , 2017

62. Costsensitive boosting for classification of imbalanced data, Mohamed S Kamel, Yang Wang, Andrew KC Wong and, Yanmin Sun, Pattern recognition, 40(12):3358–3378, , 2007

63. On the momentum term in gradient descent learning algorithms, Ning Qian, 12(1):145–151, , 1999

64. The inaturalist species classification and detection dataset, Hartwig Adam, Yin Cui, Oisin Mac Aodha, Grant Van Horn, Pietro Perona and, Yang Song, Chen Sun, Alex Shepard, Serge Belongie, In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 8769–8778, , 2018

65. The robust maximum expert consensus model with risk aversion, Ying Ji and, Yifan Ma, page 101866, 2023, , 2023

66. Adaptive class suppression loss for long-tail object detection, TongWang, Yousong Zhu, JinqiaoWang and, Wei Zeng, Chaoyang Zhao, Ming Tang, In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3103–3112, , 2021

67. Dynamic curriculum learning for imbalanced data classification, Weihao Gan, Jie Yang, Wei Wu and, Yiru Wang, Junjie Yan, In Proceedings of the IEEE/CVF international conference on computer vision, pages 5017–5026, , 2019

68. What is the effect of importance weighting in deep learning? In, Jonathon Byrd and, Zachary Lipton, International conference on machine learning, pages 872– 881. PMLR, , 2019

69. Balanced contrastive learning for long-tailed visual recognition, Yi-Ping Phoebe Chen and, Zheng Wang, Jingjing Chen, Jianggang Zhu, Yu- Gang Jiang, In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6908–6917, , 2022

70. Meta-weight-netLearning an explicit mapping for sample weighting, Qi Xie, Deyu Meng, Sanping Zhou, Lixuan Yi, Zongben Xu and, Qian Zhao, Jun Shu, Advances in neural information processing systems 32, , 2019

71. Nested collaborative learning for long-tailed visual recognition, Guodong Guo, Jun Wan, Jun Li, Zhen Lei and, Zichang Tan, In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6949–6958, , 2022

72. RsgA simple but effective module for learning imbalanced datasets, Jianfei Cai and, Xiaolin Hu, Zhenghua Xu, Jianfeng Wang, Thomas Lukasiewicz, In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3784–3793, , 2021

73. Balanced product of calibrated experts for long-tailed recognition, Michael Felsberg and, Marco Kuhlmann, Emanuel Sanchez Aimar, Arvi Jonnarth, In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19967–19977, 2023, , 1996

74. FccFeature clusters compression for long-tailed visual recognition, Rui Song, Ziyao Meng, Xiaolei Diao, Jingwen Wang and, Jian Li, Hao Xu, Daqian Shi, In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 24080–24089, 2023, , 2023

75. Subclassbalancing contrastive learning for long-tailed recognition, Haonan Wang and, Tianyi Zhou, Chengkai Hou, Jieyu Zhang, In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 5395– 5407, , 2023

76. A multiple resampling method for learning from imbalanced data sets, Andrew Estabrooks, Nathalie Japkowicz, Taeho Jo and, 20(1):18–36, , 2004

77. Disentangling label distribution for long-tailed visual recognition, Kwanghee Choi, Youngkyu Hong, Seokjun Seo, Beomsu Kim and, Seungju Han, Buru Chang, In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6626–6636, , 2021

78. Manifold MixupBetter Representations by Interpolating Hidden States, Amir Najafi, Vikas Verma, David Lopez-Paz and, Yoshua Bengio, Ioannis Mitliagkas, Alex Lamb, Christopher Beckham, In International conference on machine learning, pages 6438–6447 PMLR, , 2019

79. Retrieval augmented classification for long-tail visual recognition, Chunhua Shen and, Anton van den Hengel, Alexander Long, Thalaiyasingam Ajanthan, Ravi Garg, Alan Blair, Pulak Purkait, Vu Nguyen, Wei Yin, In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6959–6969, , 2022

80. Targeted supervised contrastive learning for longtailed recognition, Lijie Fan, Rogerio S Feris, Dina Katabi, Yuan Yuan, Yuzhe Yang, Piotr Indyk and, Peng Cao, Tianhong Li, In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6918–6928, , 2022

81. Decoupling representation and classifier for long-tailed recognition, Yannis Kalantidis, Marcus Rohrbach, Saining Xie, Albert Gordo, Jiashi Feng and, Bingyi Kang, Zhicheng Yan, arXiv preprint arXiv:1910.09217, , 2019

82. Imbalanced deep learning by minority class incremental rectification, Qi Dong, Xiatian Zhu, Shaogang Gong and, IEEE transactions on pattern analysis and machine intelligence, 41(6):1367–1381, , 2018

83. Longtailed recognition by routing diverse distribution-aware experts, Xudong Wang, Zhongqi Miao, Ziwei Liu and, Stella X Yu, Long Lian, arXiv preprint arXiv:2010.01809, , 2020

84. A Simple Framework for Contrastive Learning of Visual Representations, Ting Chen, Mohammad Norouzi and, Geoffrey Hinton, Simon Kornblith, In International conference on machine learning, pages 1597–1607. PMLR, , 2020

85. MetasaugMeta semantic augmentation for long-tailed visual recognition, Xinjing Cheng, Kaixiong Gong, Feng Qiao and, Yulin Wang, Shuang Li, Chi Harold Liu, In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5212–5221, , 2021

86. Learning imbalanced datasets with label-distribution-aware margin loss, Nikos Arechiga and, Tengyu Ma, Adrien Gaidon, ColinWei, Kaidi Cao, 32, , 2019

87. Online adaptive asymmetric active learning for budgeted imbalanced data, Peilin Zhao, Wenye Ma, Jiezhang Cao, Mingkui Tan, Qingyao Wu and, Junzhou Huang, Yifan Zhang, In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 2768–2777, , 2018

88. No one left behindImproving the worst categories in long-tailed learning, Jianxin Wu, Yingxiao Du and, In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 15804–15813, , 2023

89. AreaAdaptive reweighting via effective area for long-tailed classification, Bo Li, Chule Yang, Yucan Zhou, Xiaohua Chen, Weiping Wang, Dayan Wu, Qinghua Hu and, In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 19277–19287, 2023, , 1927

90. Distribution alignmentA unified framework for long-tail visual recognition, Jian Sun, Songyang Zhang, Zeming Li, Xuming He and, Shipeng Yan, In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2361–2370, , 2021

91. Economics of disagreement—financial intuition for the r´enyi divergence, Andrei N Soklakov, 22(8):860, , 2020

92. Feature transfer learning for face recognition with under-represented data, Manmohan Chandraker, Xi Yin, Kihyuk Sohn, Xiang Yu, Xiaoming Liu and, In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5704–5713, , 2019

93. Borderline-smotea new over-sampling method in imbalanced data sets learning, Wen-Yuan Wang and, Bing-Huan Mao, Hui Han, In Advances in Intelligent Computing: International Conference on Intelligent Computing, ICIC 2005, Hefei, China, Proceedings, Part I 1, pages 878–887. Springer, , 2005

94. C2am lossChasing a better decision boundary for long- tail object detection, Bin Yu, Yingying Chen, Ming Tang, Chaoyang Zhao, Yousong Zhu, Jinqiao Wang and, Tong Wang, In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6980–6989, , 2022

95. Curvature-balanced feature manifold learning for long-tailed classification, Xu Liu and, Lingling Li, Shuyuan Yang, Yanbiao Ma, Licheng Jiao, Fang Liu, In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15824–15835, 2023, , 2023

96. AceAlly complementary experts for solving long-tailed recognition in one-shot, Jenq-Neng Hwang, Yizhou Wang and, Jiarui Cai, In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 112–121, , 2021

97. Large scale fine-grained categorization and domain-specific transfer learning, Serge Belongie, Chen Sun, Andrew Howard and, Yin Cui, Yang Song, In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4109–4118, , 2018

98. SuperdiscoSuper-class discovery improves visual recognition for the long-tail, Cees GM Snoek, Yingjun Du, Xiantong Zhen and, Jiayi Shen, Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19944–19954, 2023, , 1994

99. Class-conditional sharpness-aware minimization for deep long-tailed recognition, Wei Gong, Pheng-Ann Heng and, Lanqing Li, Zhipeng Zhou, Peilin Zhao, In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 3499–3509, , 2023

100. A systematic study of the class imbalance problem in convolutional neural networks, Maciej A. Mazurowski, Atsuto Maki and, Mateusz Buda, 106:249–259, , 2018

101. Understanding and improving the role of projection head in selfsupervised learning, Stephen Gould, Thalaiyasingam Ajanthan, Kartik Gupta, Anton van den Hengel and, arXiv preprint arXiv:2212.11491, , 2022

102. Distributionbalanced loss for multi-label classification in long-tailed datasets In, Dahua Lin, Ziwei Liu, Yu Wang and, Tong Wu, Qingqiu Huang, Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, Proceedings, Part IV 16, pages 162–178. Springer, , 2020

103. BbnBilateralbranch network with cumulative learning for long-tailed visual recognition, Xiu-Shen Wei and, Zhao-Min Chen, Boyan Zhou, Quan Cui, In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9719–9728, , 2020

104. FasaFeature augmentation and sampling adaptation for long-tailed instance segmentation, Chen Huang and, Chen Change Loy, Yuhang Zang, In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3457– 3466, , 2021

105. Bag of tricks for long-tailed visual recognition with deep convolutional neural networks, Xiu-Shen Wei, Jianxin Wu, Boyan Zhou and, Yongshun Zhang, In Proceedings of the AAAI conference on artificial intelligence, volume 35, pages 3447–3455, , 2021

106. Self-supervised aggregation of diverse experts for test-agnostic long-tailed recognition, Bryan Hooi, Jiashi Feng, Yifan Zhang, Lanqing Hong and, Advances in Neural Information Processing Systems, 35:34077–34090, , 2022

107. Batch normalizationAccelerating deep network training by reducing internal covariate shift, Christian Szegedy, Sergey Ioffe and, In International conference on machine learning, pages 448–456. pmlr, , 2015

108. Overcoming classifier imbalance for long-tail object detection with balanced group softmax, Jintao Li and, Bingyi Kang, Sheng Tang, ChunfengWang, Yu Li, TaoWang, Jiashi Feng, In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10991–11000, , 2020

109. knn approach to unbalanced data distributionsa case study involving information extraction, I Zhang, Inderjeet Mani and, In Proceedings of workshop on learning from imbalanced datasets, volume 126, pages 1–7. ICML, , 2003

110. Long-tailed visual recognition via self-heterogeneous integration with knowledge excavation, Yiu-ming Cheung and, HanziWang, Mengke Li, Yang Lu, Yan Jin, In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 23695–23704, 2023, , 2023

111. Training cost-sensitive neural networks with methods addressing the class imbalance problem, Xu-Ying Liu, Zhi-Hua Zhou and, IEEE Transactions on knowledge and data engineering, 18(1):63–77, , 2005

112. and MdcsMore diverse experts with consistency self-distillation for long-tailed recognition, Chen Jiang, Qihao Zhao, Wei Hu, Jun Liu, Fan Zhang, In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 11597–11608, , 2023

113. Vl-ltrLearning class-wise visual-linguistic representation for long-tailed visual recognition, Changyao Tian, Yu Qiao, Wenhai Wang, Xizhou Zhu, Jifeng Dai and, In Computer Vision–ECCV 2022: 17th European Conference, Tel Aviv, Israel Proceedings, Part XXV, pages 73–91. Springer, , 2022

114. Deep representation learning on long-tailed dataA learnable embedding augmentation perspective, Chuchu Han, Wenhui Li, Jialun Liu, Yifan Sun, Zhaopeng Dou and, In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2970–2979, , 2020

115. Global and local mixture consistency cumulative learning for long-tailed visual recog- nitions, Yun Yang, Peng Yang, Qi Jia, Fengtao Nan, Fei Du, Xiaoting Chen and, Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15814–15823, 2023, , 2023

116. Learning from multiple expertsSelf-paced knowledge distillation for long-tailed classification, Liuyu Xiang, Jungong Han, Guiguang Ding and, In Com- puter Vision–ECCV 2020: 16th European Conference, Glasgow, UK, Proceedings, Part V 16, pages 247–263. Springer, , 2020

117. Consensus-oriented decision-makingThe CODM model for facilitating groups to widespread agreement, Tim Hartnett, new society publishers, , 2011

118. The majority can help the minorityContext-rich minority oversampling for long-tailed classification, Youngkyu Hong, Jin Young Choi, Byeongho Heo, Seulki Park, Sangdoo Yun and, In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6887–6896, , 2022

119. Long-tailed multi-label visual recognition by collaborative training on uniform and re-balanced samplings In, Song Wang, Hao Guo and, Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15089–15098, , 2021

120. Safe-level-smoteSafe-level-synthetic minority over-sampling technique for handling the class imbalanced problem, Chidchanok Lursinsap, Chumphol Bunkhumpornpat, Krung Sinapiromsaran and, In Advances in Knowledge Discovery and Data Mining: 13th Pacific-Asia Conference, PAKDD 2009 Bangkok, Thailand, pages 475–482 Springer, , 2009

121. AdaboostcnnAn adaptive boosting algorithm for convolutional neural networks to classify multi-class imbalanced datasets using transfer learning, Aboozar Taherkhani, T Martin McGinnity, Georgina Cosma and, 404:351–366, , 2020