Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-supervised Learning is More Robust to Dataset Imbalance
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use self-supervised rather than supervised pre-training | MoCo v2 versus supervised learning · ImageNet ID and OOD evaluations across imbalance configurations | relative accuracy gap between balanced and imbalanced pre-training | Helped |
| Use self-supervised rather than supervised pre-training | MoCo v2 versus supervised learning · ImageNet-LT, semi-synthetic rare-class evaluation | accuracy: 70.1% vs $44.3\%$ | Helped |
| Add density-weighted SAM | SimSiam · CIFAR-10-LT step, r = 0.01, ID | accuracy: 84.3 $\pm$ 0.2 -> 85.6 $\pm$ 0.4 | Helped |
| Add density-weighted SAM | SimSiam · CIFAR-10-LT exponential, r = 0.01, ID | accuracy: 81.4 $\pm$ 0.3 -> 82.7 $\pm$ 0.5 | Helped |
| Add ordinary SAM without density reweighting (ablation) | SimSiam · CIFAR-10-LT step, r = 0.01, ID | accuracy: 84.3 $\pm$ 0.2 -> 84.7 $\pm$ 0.3 | Helped |
| Add ordinary SAM without density reweighting (ablation) | SimSiam · CIFAR-10-LT exponential, r = 0.01, ID | accuracy: 81.4 $\pm$ 0.3 -> 82.1 $\pm$ 0.4 | Helped |
| Replace ordinary SAM with density-weighted SAM (ablation) | SimSiam · CIFAR-10-LT step, r = 0.01, ID | accuracy: 84.7 $\pm$ 0.3 -> 85.6 $\pm$ 0.4 | Helped |
| Replace ordinary SAM with density-weighted SAM (ablation) | SimSiam · CIFAR-10-LT exponential, r = 0.01, ID | accuracy: 82.1 $\pm$ 0.4 -> 82.7 $\pm$ 0.5 | Helped |
| Use density-weighted SAM | MoCo v2 · ImageNet-LT downstream average OOD | average accuracy: 80.3 -> 81.4 | Helped |
| Use density-weighted SAM | SimSiam · ImageNet-LT downstream average OOD | average accuracy: 80.6 -> 81.4 | Helped |
| Use ordinary SAM without density reweighting (ablation) | MoCo v2 · ImageNet-LT downstream average OOD | average accuracy: 80.3 -> 80.9 | Helped |
| Replace ordinary SAM with density-weighted SAM (ablation) | MoCo v2 · ImageNet-LT downstream average OOD | average accuracy: 80.9 -> 81.4 | Helped |
Closest papers by meaning
- Rethinking the Value of Labels for Improving Class-Imbalanced Learning (NeurIPS 2020 · similarity 0.91)
- Exploring Balanced Feature Spaces for Representation Learning (ICLR 2021 · similarity 0.90)
- Views Can Be Deceiving: Improved SSL Through Feature Space Augmentation (ICLR 2024 · similarity 0.88)
- Robust Visual Recognition with Class-Imbalanced Open-World Noisy Data (AAAI 2024 · similarity 0.88)
- BaCon: Boosting Imbalanced Semi-supervised Learning via Balanced Feature-Level Contrastive Learning (AAAI 2024 · similarity 0.87)
- Measuring Self-Supervised Representation Quality for Downstream Classification Using Discriminative Features (AAAI 2024 · similarity 0.87)
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.