Boosting Few-Shot Learning With Adaptive Margin Loss

By Aoxue Li, Weiran Huang, Xu Lan, Jiashi Feng, Zhenguo Li, Liwei Wang. 2020

224 citations recorded by the source corpus.

Few-shot learning (FSL) has attracted increasing attention in recent years but remains challenging, due to the intrinsic difficulty in learning to generalize from a few examples. This paper proposes an adaptive margin principle to improve the generalization ability of metric-based meta-learning approaches for few-shot learning problems. Specifically, we first develop a class-relevant additive margin loss, where semantic similarity between each pair of classes is considered to separate samples in the feature embedding space from similar classes. Further, we incorporate the semantic context among all classes in a sampled training task and develop a task-relevant additive margin loss to better distinguish samples from different classes. Our adaptive margin method can be easily extended to a more realistic generalized FSL setting. Extensive experiments demonstrate that the proposed method can boost the performance of current metric-based meta-learning approaches, under both the standard FSL and generalized FSL settings.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

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. Hanwang Zhang
2. Xiansheng Hua
3. Yunchao Wei
4. Yi Yang
5. Qianru Sun
6. Qixiang Ye
7. Jianbin Jiao
8. Shanghang Zhang

In Computer Vision and Pattern Recognition

1. Yi Yang
2. Hanwang Zhang
3. Dacheng Tao
4. Luc Van Gool
5. Yunchao Wei
6. Bernt Schiele
7. Xiansheng Hua
8. Qianru Sun

Citations in the corpus

• Margin-Based Few-Shot Class-Incremental Learning with Class-Level Overfitting Mitigation (2022, methodology): “(2) Relational-margin-based methods (TRAML [20]), which maps the class relationship to the margin.”
• Interventional Few-Shot Learning (2020, background): “6 [34] Aoxue Li, Weiran Huang, Xu Lan, Jiashi Feng, Zhenguo Li, and Liwei Wang.”
• Meta-Baseline: Exploring Simple Meta-Learning for Few-Shot Learning (2020, unknown): “47 ProtoNets + TRAML [11] ResNet-12 60.”
• Anti-aliasing Semantic Reconstruction for Few-Shot Semantic Segmentation (2021, background): “Extra margin constraints [20, 17] are absorbed into metric learning to further adjust the inter-class diversity and intra-class variance.”
• Orthogonal Projection Loss (2021, unknown): “Note that [40, 28, 31] are recent loss functions specific to FSL. Dataset CNAPs [42] (NeurIPS’19) SUR [11] (ECCV’20”
• Consistent Structural Relation Learning for Zero-Shot Segmentation (2020, background): “To extremely reduce the cost of label annotation, previous works focus on weakly-supervised segmentation [7, 8, 43] and few-shot segmentation [17, 44].”
• A Comprehensive Survey of Few-shot Learning: Evolution, Applications, Challenges, and Opportunities (2022, background): “BF3S [175] CVPR’20 Episodic Training Cosine Similarity Adaptive boundary distances between similar categories larger than between dissimilar categories”
• Curriculum-Based Meta-learning (2021, methodology): “Instead of evaluating the models on only 600 (or 1000) randomly sampled testing tasks (15 query samples per class) [1, 21, 27], we evaluate methods on 3000 sampled tasks for lower variances.”