Research map / Label-efficient and robust learning / Self-supervised and contrastive
Improving Self-supervised Learning with Automated Unsupervised Outlier Arbitration
Yu Wang, Jingyang Lin, Jingjing Zou, Yingwei Pan, Ting Yao, Tao Mei
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 UOTA | MoCo-v2 · ImageNet100 | Top-1: 73.0 -> 74.0 | Improved |
| add UOTA | MoCo-v2* · ImageNet100 | Top-1: 80.3 -> 81.4 | Improved |
| add UOTA | BYOL · ImageNet100 | Top-1: 75.3 -> 76.2 | Improved |
| add UOTA | SwAV · ImageNet100 | Top-1: 81.1 -> 82.2 | Improved |
| add UOTA | SwAV ResNet-50 · ImageNet1K | Top-1: 72.7 -> 73.5 | Improved |
| add focal loss | MoCo-v2* · ImageNet100 | Top-1: 80.3 -> 79.6 | Hurt |
| add GCE, MIL-NCE or Debiased | MoCo-v2* · ImageNet100 | Top-1: 80.3 -> 80.8 | Mixed |
| use UOTA instead of Debiased | MoCo-v2* · ImageNet100 | Top-1: 80.7 -> 81.4 | Improved |
| replace global covariance with identity (no whitening) (ablation) | MoCo-v2*+UOTA · ImageNet100 | Top-1: 81.4 -> 80.7 | Hurt |
| replace global covariance with local per-instance covariance (ablation) | MoCo-v2*+UOTA · ImageNet100 | Top-1: 81.4 -> 80.7 | Hurt |
| use original feature z_i instead of average mu_i as mean (ablation) | MoCo-v2*+UOTA · ImageNet100 | Top-1: 81.4 -> 81.3 | No clear effect |
| UOTA instead of NDA | MoCo-v2 · ImageNet100 ResNet-50 | Top-1: 70.0 -> 70.6 | Improved |
Closest papers by meaning
- Using Self-Supervised Learning Can Improve Model Robustness and Uncertainty (NeurIPS 2019 · similarity 0.87)
- Self-Joint Supervised Learning (ICLR 2022 · similarity 0.87)
- On the Out-of-Distribution Generalization of Self-Supervised Learning (ICML 2025 · similarity 0.86)
- Self-supervised Learning is More Robust to Dataset Imbalance (ICLR 2022 · similarity 0.86)
- ODIM: Outlier Detection via Likelihood of Under-Fitted Generative Models (ICML 2024 · similarity 0.85)
- ALTBI: Constructing Improved Outlier Detection Models via Optimization of Inlier-Memorization Effect (AAAI 2025 · similarity 0.84)
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.
