Research map / Label-efficient and robust learning / Recognition and long-tail
Improving Contrastive Learning on Imbalanced Data via Open-World Sampling
Ziyu Jiang, Tianlong Chen, Ting Chen, Zhangyang Wang
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 |
|---|---|---|---|
| Use MAK to select 10K external images | SimCLR, ResNet-50 on ImageNet-100-LT · ImageNet-900; linear separability | All accuracy: 71.2±0.2 -> 72.7±0.4 | Improved |
| Use MAK rather than random sampling | SimCLR, ResNet-50 on ImageNet-100-LT · ImageNet-900; linear separability | Std (imbalancedness): 3.5±0.5 -> 3.0±0.1 | Improved |
| Use MAK to select 10K external images | SimCLR, ResNet-50 on ImageNet-100-LT · ImageNet-900; few-shot | All accuracy: 51.1±0.1 -> 51.9±0.4 | Improved |
| Use MAK rather than random sampling | SimCLR, ResNet-50 on ImageNet-100-LT · ImageNet-Places-Mix; linear separability | All accuracy: 69.8±0.5 -> 71.1±0.5 | Improved |
| Use MAK rather than random sampling | SimCLR, ResNet-50 on ImageNet-100-LT · ImageNet-Places-Mix; few-shot | All accuracy: 48.7±0.4 -> 49.3±0.7 | Improved |
| Add tailness selection alone (ablation) | MAK sampling on ImageNet-100-LT · ImageNet-900; linear separability | All accuracy: 71.2±0.2 -> 70.9±0.4 | Hurt |
| Add proximity selection alone (ablation) | MAK sampling on ImageNet-100-LT · ImageNet-900; linear separability | All accuracy: 71.2±0.2 -> 70.2±0.4 | Hurt |
| Add diversity selection alone (ablation) | MAK sampling on ImageNet-100-LT · ImageNet-900; linear separability | All accuracy: 71.2±0.2 -> 70.0±0.4 | Hurt |
| Combine tailness and proximity (ablation) | MAK sampling on ImageNet-100-LT · ImageNet-900; linear separability | All accuracy: 71.2±0.2 -> 72.2±0.2 | Improved |
| Add diversity to tailness and proximity (ablation) | MAK sampling on ImageNet-100-LT · ImageNet-900; linear separability | All accuracy: 72.2±0.2 -> 72.7±0.4 | Improved |
| Remove tailness from full MAK (ablation) | MAK sampling on ImageNet-100-LT · ImageNet-900; few-shot | All accuracy: 51.9±0.4 -> 49.1±0.3 | Hurt |
| Remove proximity from full MAK (ablation) | MAK sampling on ImageNet-100-LT · ImageNet-900; few-shot | All accuracy: 51.9±0.4 -> 49.3±0.4 | Hurt |
Closest papers by meaning
- A Tale of Two Classes: Adapting Supervised Contrastive Learning to Binary Imbalanced Datasets (CVPR 2025 · similarity 0.88)
- Robust Visual Recognition with Class-Imbalanced Open-World Noisy Data (AAAI 2024 · similarity 0.88)
- ConR: Contrastive Regularizer for Deep Imbalanced Regression (ICLR 2024 · similarity 0.87)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 2025 · similarity 0.87)
- Continuous Contrastive Learning for Long-Tailed Semi-Supervised Recognition (NeurIPS 2024 · similarity 0.86)
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 2024 · similarity 0.86)
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.
