Generalized Adaptation for Few-Shot Learning
2019-11-25Unverified0· sign in to hype
Liang Song, Jinlu Liu, Yongqiang Qin
Unverified — Be the first to reproduce this paper.
ReproduceAbstract
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%.
Tasks
Benchmark Results
| Dataset | Model | Metric | Claimed | Verified | Status |
|---|---|---|---|---|---|
| CIFAR-FS 5-way (1-shot) | ACC + Amphibian | Accuracy | 73.1 | — | Unverified |
| CIFAR-FS 5-way (5-shot) | ACC + Amphibian | Accuracy | 89.3 | — | Unverified |
| FC100 5-way (1-shot) | ACC + Amphibian | Accuracy | 41.6 | — | Unverified |
| FC100 5-way (5-shot) | ACC + Amphibian | Accuracy | 66.9 | — | Unverified |
| Mini-Imagenet 5-way (1-shot) | ACC + Amphibian | Accuracy | 62.21 | — | Unverified |
| Mini-Imagenet 5-way (5-shot) | ACC + Amphibian | Accuracy | 80.75 | — | Unverified |
| Tiered ImageNet 5-way (1-shot) | ACC + Amphibian | Accuracy | 68.77 | — | Unverified |
| Tiered ImageNet 5-way (5-shot) | ACC + Amphibian | Accuracy | 86.75 | — | Unverified |