Research map / Label-efficient and robust learning / Recognition and long-tail
Neighborhood Contrastive Learning for Novel Class Discovery
Check 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 |
|---|---|---|---|
| Add neighborhood contrastive learning with pseudo-positives | Baseline · CIFAR-10 | ACC: 87.9$\pm$0.7\% -> 93.4$\pm$0.2\% | Helped |
| Add neighborhood contrastive learning with pseudo-positives | Baseline · CIFAR-100 | ACC: 69.4$\pm$1.4\% -> 82.3$\pm$2.6\% | Helped |
| Add hard negative generation | NCL · CIFAR-100 | ACC: 82.3$\pm$2.6\% -> 86.6$\pm$0.4\% | Helped |
| Add hard negative generation | NCL · CIFAR-10 | ACC: 93.4$\pm$0.2\% -> 93.4$\pm$0.1\% | No clear effect |
| Add hard negative generation | NCL · ImageNet | ACC: 89.5\% -> 90.7\% | Helped |
| Remove pseudo-positives from NCL (ablation) | NCL · CIFAR-10 | ACC: 93.4$\pm$0.2\% -> 61.8$\pm$7.6\% | Hurt |
| Remove pseudo-positives from NCL (ablation) | NCL · CIFAR-100 | ACC: 82.3$\pm$2.6\% -> 68.5$\pm$1.9\% | Hurt |
| Remove the augmented-positive (ablation) | NCL · CIFAR-10 | ACC: 93.4$\pm$0.2\% -> 90.9$\pm$2.1\% | Hurt |
| Remove the augmented-positive (ablation) | NCL · CIFAR-100 | ACC: 82.3$\pm$2.6\% -> 79.7$\pm$5.7\% | Hurt |
| Remove NCL on labeled data (ablation) | NCL · CIFAR-100 | ACC: 82.3$\pm$2.6\% -> 80.3$\pm$0.9\% | Hurt |
| Remove NCL on labeled data (ablation) | NCL · CIFAR-10 | ACC: 93.4$\pm$0.2\% -> 93.3$\pm$0.1\% | No clear effect |
| Remove self-supervised learning (ablation) | Baseline · CIFAR-10 | ACC: 87.9$\pm$0.7\% -> 85.0$\pm$0.4\% | Hurt |
Closest papers by meaning
- Modeling Inter-Class and Intra-Class Constraints in Novel Class Discovery (CVPR 2023 · similarity 0.88)
- When and How Does Known Class Help Discover Unknown Ones? Provable Understanding Through Spectral Analysis (ICML 2023 · similarity 0.88)
- Debiased Novel Category Discovering and Localization (AAAI 2024 · similarity 0.86)
- Semantic-Guided Novel Category Discovery (AAAI 2024 · similarity 0.86)
- Self-Supervised Contrastive Learning is Approximately Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.85)
- Non-negative Contrastive Learning (ICLR 2024 · similarity 0.85)
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.