Research map / Label-efficient and robust learning / Noisy labels
Robust Dataset Condensation using Supervised Contrastive Learning
Nicole Hee-Yeon Kim, Hwanjun Song
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 RDC (golden MixUp contrast + SSL) | IDM · CIFAR-10 asymmetric 40% noise, 10 img/cls | test accuracy (%): 30.61 -> 41.35 | Improved |
| add RDC | IDM · CIFAR-10 asymmetric 40% noise, 50 img/cls | test accuracy (%): 34.85 -> 55.55 | Improved |
| add RDC | IDM · CIFAR-100 clean, 10 img/cls | test accuracy (%): 18.94 -> 26.73 | Improved |
| add RDC instead of two-stage cleaning | IDM · CIFAR-100 symmetric 40% noise, 10 img/cls | test accuracy (%): 19.85 -> 17.60 | Hurt |
| add RDC instead of two-stage cleaning | IDM · CIFAR-10 asymmetric 40% noise, 10 img/cls | test accuracy (%): 33.29 -> 41.35 | Improved |
| add two-stage DivideMix cleaning before condensation | IDM · Tiny-ImageNet asymmetric 40% noise, 10 img/cls | test accuracy (%): 6.38 -> 0.51 | Hurt |
| add RDC | Acc-DD · CIFAR-10 asymmetric 40% noise, 10 img/cls | test accuracy (%): 35.33 -> 55.18 | Improved |
| add RDC | Acc-DD · CIFAR-10 clean, 10 img/cls | test accuracy (%): 57.18 -> 60.99 | Improved |
| add semi-supervised learning (DivideMix) using golden set (ablation) | IDM · CIFAR-10 asymmetric 40% noise, 10 img/cls | test accuracy (%): 30.61 -> 35.23 | Improved |
| add SupCon on synthetic set only (no augmentation) (ablation) | IDM + SSL · CIFAR-10 asymmetric 40% noise, 10 img/cls | test accuracy (%): 35.23 -> 33.89 | Hurt |
| add simple flip/crop augmentation to SupCon (ablation) | IDM + SSL + SupCon · CIFAR-10 asymmetric 40% noise, 10 img/cls | test accuracy (%): 33.89 -> 35.24 | No clear effect |
| add golden MixUp contrast (ablation) | IDM + SSL + SupCon w. augment · CIFAR-10 asymmetric 40% noise, 10 img/cls | test accuracy (%): 35.24 -> 41.35 | Improved |
Closest papers by meaning
- Dataset Condensation with Contrastive Signals (ICML 2022 · similarity 0.89)
- Elucidating the Design Space of Dataset Condensation (NeurIPS 2024 · similarity 0.87)
- Ameliorate Spurious Correlations in Dataset Condensation (ICML 2024 · similarity 0.86)
- Diversity-Driven Synthesis: Enhancing Dataset Distillation through Directed Weight Adjustment (NeurIPS 2024 · similarity 0.84)
- Multisize Dataset Condensation (ICLR 2024 · similarity 0.84)
- Towards Syn-to-Real IQA: A Novel Perspective on Reshaping Synthetic Data Distributions (NeurIPS 2025 · similarity 0.83)
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.
