Rethinking the Value of Labels for Improving Class-Imbalanced Learning

By Yuzhe Yang, Zhi Xu. 2020

499 citations recorded by the source corpus.

Real-world data often exhibits long-tailed distributions with heavy class imbalance, posing great challenges for deep recognition models. We identify a persisting dilemma on the value of labels in the context of imbalanced learning: on the one hand, supervision from labels typically leads to better results than its unsupervised counterparts; on the other hand, heavily imbalanced data naturally incurs "label bias" in the classifier, where the decision boundary can be drastically altered by the majority classes. In this work, we systematically investigate these two facets of labels. We demonstrate, theoretically and empirically, that class-imbalanced learning can significantly benefit in both semi-supervised and self-supervised manners. Specifically, we confirm that (1) positively, imbalanced labels are valuable: given more unlabeled data, the original labels can be leveraged with the extra data to reduce label bias in a semi-supervised manner, which greatly improves the final classifier; (2) negatively however, we argue that imbalanced labels are not useful always: classifiers that are first pre-trained in a self-supervised manner consistently outperform their corresponding baselines. Extensive experiments on large-scale imbalanced datasets verify our theoretically grounded strategies, showing superior performance over previous state-of-the-arts. Our intriguing findings highlight the need to rethink the usage of imbalanced labels in realistic long-tailed tasks. Code is available at this https URL.

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

Citations in the corpus

S3C: Self-Supervised Stochastic Classifiers for Few-Shot Class-Incremental Learning (2023, methodology): “…performance of downstream tasks like few-shot learning [13], semi-supervised learning [43], to improve the model robustness [14], class imbalance [40], etc. Recently, Lee et al. [22] used SSL to improve the performance for supervised classification, by augmenting the original labels using the…”
Prototype Augmentation and Self-Supervision for Incremental Learning (2021, background): “In particular, it has been found that self-supervised pretraining is a good choice to initialize the model for class-imbalance learning [52].”
Disentangling Label Distribution for Long-tailed Visual Recognition (2020, background): “Both strategies suffer from under-representation of head classes or memorization of tail classes [62, 9].”
FreeMatch: Self-adaptive Thresholding for Semi-supervised Learning (2022, methodology): “Inspired by [26], we consider a binary classification problem where the true distribution is an even mixture of two Gaussians (i.”
A Survey on Self-Supervised Learning: Algorithms, Applications, and Future Trends (2023, unknown): “Several studies have delved into this area [54], [71], [72], [73], [74], [75], [76], [77], [78], [79], [80], [81].”
Self-Supervised Learning of Graph Neural Networks: A Unified Review (2021, methodology): “Instead of the labels obtained from input graphs, the prediction targets in self-training methods are pseudo-labels obtained from the prediction in a previous stage utilizing a small portion of labeled data [1] or even randomly initial-”
CReST: A Class-Rebalancing Self-Training Framework for Imbalanced Semi-Supervised Learning (2021, background): “Apart from a few recent works [22, 48], the majority of existing SSL algorithms [2, 1, 46, 39] have not been thoroughly evaluated on imbalanced data distributions.”
Delving into Deep Imbalanced Regression (2021, unknown): “Recent studies have also discovered that semi-supervised learning and self-supervised learning lead to better imbalanced classiﬁcation results (Yang & Xu, 2020).”
Exploring Balanced Feature Spaces for Representation Learning (2021, methodology): “Along this direction, SSP (Yang & Xu, 2020) is among the ﬁrst methods that introduce SSL pretraining into learning the long-tailed recognition models.”