Research map / Label-efficient and robust learning / Recognition and long-tail
When and How Does Known Class Help Discover Unknown Ones? Provable Understanding Through Spectral Analysis
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 NSCL instead of ComEx | C10-5 · C10-5 training novel split | clustering accuracy (%): 93.6 -> 97.5 | Helped |
| Use NSCL instead of ComEx | C100-80 · C100-80 training novel split | clustering accuracy (%): 85.7 -> 85.9 | Helped |
| Use NSCL instead of ComEx | C100-50 · C100-50 training novel split | clustering accuracy (%): 53.4 -> 64.0 | Helped |
| Add known labeled data and its label supervision to unlabeled-only SCL (ablation) | NSCL · C100-50 training novel split | clustering accuracy (%): +13.2% | Helped |
| Add label supervision for the labeled data, compared with SCL‡ (ablation) | NSCL · C100-80 training novel split | clustering accuracy (%): 68.9 -> 85.9 | Helped |
| Add label supervision for the labeled data, compared with SCL‡ (ablation) | NSCL · C100-50 training novel split | clustering accuracy (%): 53.3 -> 64.0 | Helped |
| Add label supervision for the labeled data, compared with SCL‡ (ablation) | NSCL · C10-5 training novel split | clustering accuracy (%): 93.7 -> 97.5 | Helped |
| Add a known class connected to the target novel attribute (red cylinder rather than unrelated gray cylinder) (ablation) | NCD spectral representation · Toy example, color label vector | regression residual: 1 -> 0 | Helped |
| Add known data with a stronger connection in a spurious feature dimension (gray cube) (ablation) | NCD spectral representation · Toy example, color label vector | regression residual: 0 -> 1 | Hurt |
| Increase the labeled-to-novel connection strength beyond the threshold (ablation) | NCD spectral representation · Toy example, color label vector | regression residual: r(t) -> 0 | Helped |
Closest papers by meaning
- Joint Class-level and Instance-level Relationship Modeling for Novel Class Discovery (AAAI 2025 · similarity 0.89)
- Adaptive Discovering and Merging for Incremental Novel Class Discovery (AAAI 2024 · similarity 0.85)
- Towards Understanding Parametric Generalized Category Discovery on Graphs (ICML 2025 · similarity 0.85)
- Semantic-Guided Novel Category Discovery (AAAI 2024 · similarity 0.85)
- Seeing Unseen: Discover Novel Biomedical Concepts via Geometry-Constrained Probabilistic Modeling (CVPR 2024 · similarity 0.85)
- Nonparametric Identification of Latent Concepts (ICML 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.