Advances in Long-Tail Learning

Deep learning has shown its power in the area of computer vision in recent years. Most of the deep learning models are built upon training data with balanced class distribution. However, the training data of computer vision tasks in real-world scenarios usually exhibit long-tail distribution, where a few head classes possess a large number of samples while all the rest of tail classes only possess a small number of samples. For example, the distribution of objects in daily life is highly long-tailed, and the deep learning based object detection model trained on this kind of long-tailed data poorly generalizes on the rare objects. Therefore, the research problem of long-tail learning is to propose deep learning algorithms to address the long-tail challenge for computer vision in practice.

This Research Topic aims to present novel solutions to these challenging problems and to push the frontiers of long-tail learning from both a theoretical and practical perspective. Contributions are welcome on new theories, methods, frameworks, and models, in addition to new benchmark datasets with long-tail distribution, and open source software. The Research Topic also encourages submissions on building connections with related deep learning areas such as few-shot learning, noisy-label learning, adversarial learning, continual learning, and so on. Submitted manuscripts should describe high-quality, original and innovative work that has neither appeared in, nor be under consideration by other journals and conferences.

Topics of interest include, but are not limited to:
- Long-tailed representation learning
- Long-tailed continual learning
- Long-tail learning with noisy labels
- Privacy-preserving long-tail learning
- Ensemble methods for long-tail learning
- Semi-supervised and unsupervised long-tail learning
- Long-tailed contrastive learning
- Data augmentation for long-tail learning
- Long-tail learning for specific computer vision tasks

Keywords: machine learning, deep learning, long-tail learning, data imbalance, representation learning

Important Note: All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.