Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-Supervision Can Be a Good Few-Shot Learner
Yuning Lu, Liangjian Wen, Jianzhuang Liu, Yajing Liu, Xinmei Tian
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add self-supervised knowledge distillation from ResNet-50 teacher (ablation) | UniSiam ResNet-18 · mini-ImageNet 5-way 5-shot | accuracy (%): 79.11 -> 80.35 | Improved |
| add self-supervised knowledge distillation from ResNet-50 teacher (ablation) | UniSiam ResNet-34 · mini-ImageNet 5-way 5-shot | accuracy (%): 79.69 -> 81.39 | Improved |
| add self-supervised knowledge distillation from ResNet-50 teacher (ablation) | UniSiam ResNet-10 · mini-ImageNet 5-way 5-shot | accuracy (%): 76.94 -> 78.58 | Improved |
| add uniformity regularization (UniSiam) to SimSiam under strong augmentation (ablation) | SimSiam ResNet-50 · mini-ImageNet 5-way 5-shot | accuracy (%): 79.66 -> 83.18 | Improved |
| switch from default to strong data augmentation (ablation) | SimSiam ResNet-50 · mini-ImageNet 5-way 5-shot | accuracy (%): 81.75 -> 79.66 | Hurt |
| switch from default to strong data augmentation (ablation) | UniSiam ResNet-50 · mini-ImageNet 5-way 5-shot | accuracy (%): 81.93 -> 83.18 | Improved |
| use MINE uniformity instead of NCE uniformity (ablation) | UniSiam ResNet-18 strong DA · mini-ImageNet 5-way 5-shot | accuracy (%): 80.66 -> 81.13 | Improved |
| use asymmetric alignment instead of symmetric alignment (ablation) | UniSiam ResNet-18 strong DA · mini-ImageNet 5-way 5-shot | accuracy (%): 80.72 -> 81.13 | Improved |
| increase backbone from ResNet-10 to ResNet-50 (ablation) | supervised pre-training · mini-ImageNet | accuracy (%): decreased by 0.2% | Hurt |
| increase backbone from ResNet-10 to ResNet-50 (ablation) | self-supervised pre-training · mini-ImageNet | accuracy (%): improvement larger than 4% | Improved |
| increase input image size from 160x160 to 384x384 (ablation) | self-supervised pre-training · mini-ImageNet | accuracy (%) | Improved |
| add RandAugment to default plus vertical flip augmentation (ablation) | SimSiam · mini-ImageNet | accuracy (%) | No clear effect |
Closest papers by meaning
- Boosting Few-Shot Visual Learning With Self-Supervision (ICCV 2019 · similarity 0.93)
- When Does Self-supervision Improve Few-shot Learning? (ECCV 2020 · similarity 0.89)
- Self-supervised Learning from a Multi-view Perspective (ICLR 2021 · similarity 0.89)
- Learning to Self-Train for Semi-Supervised Few-Shot Classification (NeurIPS 2019 · similarity 0.89)
- Pareto Self-Supervised Training for Few-Shot Learning (CVPR 2021 · similarity 0.89)
- Self-Supervised Learning via Maximum Entropy Coding (NeurIPS 2022 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
