Liang Song, Jinlu Liu, Yongqiang Qin
Many Few-Shot Learning research works have two stages: pre-training base model and adapting to novel model. In this paper, we propose to use closed-form base learner, which constrains the adapting stage with pre-trained base model to get better generalized novel model. Following theoretical analysis proves its rationality as well as indication of how to train a well-generalized base model. We then conduct experiments on four benchmarks and achieve state-of-the-art performance in all cases. Notably, we achieve the accuracy of 87.75% on 5-shot miniImageNet which approximately outperforms existing methods by 10%.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Image Classification | CIFAR-FS 5-way (1-shot) | Accuracy | 73.1 | ACC + Amphibian |
| Image Classification | Mini-Imagenet 5-way (5-shot) | Accuracy | 80.75 | ACC + Amphibian |
| Image Classification | Mini-Imagenet 5-way (1-shot) | Accuracy | 62.21 | ACC + Amphibian |
| Image Classification | FC100 5-way (5-shot) | Accuracy | 66.9 | ACC + Amphibian |
| Image Classification | FC100 5-way (1-shot) | Accuracy | 41.6 | ACC + Amphibian |
| Image Classification | Tiered ImageNet 5-way (1-shot) | Accuracy | 68.77 | ACC + Amphibian |
| Image Classification | Tiered ImageNet 5-way (5-shot) | Accuracy | 86.75 | ACC + Amphibian |
| Image Classification | CIFAR-FS 5-way (5-shot) | Accuracy | 89.3 | ACC + Amphibian |
| Few-Shot Image Classification | CIFAR-FS 5-way (1-shot) | Accuracy | 73.1 | ACC + Amphibian |
| Few-Shot Image Classification | Mini-Imagenet 5-way (5-shot) | Accuracy | 80.75 | ACC + Amphibian |
| Few-Shot Image Classification | Mini-Imagenet 5-way (1-shot) | Accuracy | 62.21 | ACC + Amphibian |
| Few-Shot Image Classification | FC100 5-way (5-shot) | Accuracy | 66.9 | ACC + Amphibian |
| Few-Shot Image Classification | FC100 5-way (1-shot) | Accuracy | 41.6 | ACC + Amphibian |
| Few-Shot Image Classification | Tiered ImageNet 5-way (1-shot) | Accuracy | 68.77 | ACC + Amphibian |
| Few-Shot Image Classification | Tiered ImageNet 5-way (5-shot) | Accuracy | 86.75 | ACC + Amphibian |
| Few-Shot Image Classification | CIFAR-FS 5-way (5-shot) | Accuracy | 89.3 | ACC + Amphibian |