Boosting Few-Shot Classification with View-Learnable Contrastive Learning
This addresses the challenge of fine-grained generalization in few-shot classification, which is incremental as it builds on existing metric-based methods.
The paper tackles the problem of discriminating fine-grained sub-categories in few-shot classification by introducing contrastive loss and a learning-to-learn algorithm for view generation, achieving superior performance on standard benchmarks.
The goal of few-shot classification is to classify new categories with few labeled examples within each class. Nowadays, the excellent performance in handling few-shot classification problems is shown by metric-based meta-learning methods. However, it is very hard for previous methods to discriminate the fine-grained sub-categories in the embedding space without fine-grained labels. This may lead to unsatisfactory generalization to fine-grained subcategories, and thus affects model interpretation. To tackle this problem, we introduce the contrastive loss into few-shot classification for learning latent fine-grained structure in the embedding space. Furthermore, to overcome the drawbacks of random image transformation used in current contrastive learning in producing noisy and inaccurate image pairs (i.e., views), we develop a learning-to-learn algorithm to automatically generate different views of the same image. Extensive experiments on standard few-shot learning benchmarks demonstrate the superiority of our method.