Research map / Continual learning / Class-incremental
Proxy Anchor-based Unsupervised Learning for Continuous Generalized Category Discovery
Hyungmin Kim, Sungho Suh, Daehwan Kim, Daun Jeong, Hansang Cho, Junmo Kim
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 the fine split (noisy label learning) after the cosine-similarity initial split (ablation) | CGCD framework · CUB-200 (ResNet-18) | M_d: 36.47 -> 40.90 | Improved |
| add the fine split with ViT-B-16 backbone (ablation) | CGCD framework · CUB-200 (ViT-B-16) | M_all: 70.22 -> 72.51 | Improved |
| add proxy-anchor-based exemplar feature replay (ablation) | CGCD framework · CUB-200 (ResNet-18) | M_f (lower is better): 36.88 -> 15.47 | Improved |
| add proxy-anchor-based exemplar (vs none) (ablation) | CGCD framework · CUB-200 (ResNet-18) | M_d: 41.89 -> 40.90 | Hurt |
| use proxy-anchor exemplar instead of data mean and std exemplar (ablation) | CGCD framework · CUB-200 (ResNet-18) | M_all: 40.23 -> 54.75 | Improved |
| use proxy anchor loss instead of ArcFace (ablation) | CGCD framework · CUB-200 (ResNet-18) | M_d: 28.02 -> 40.90 | Improved |
| use proxy anchor loss instead of ProxyNCA++ (ablation) | CGCD framework · CUB-200 (ResNet-18) | M_f (lower is better): 8.85 -> 15.47 | Hurt |
| reduce the labeled fraction of classes (old:new from 9:1 to 5:5) (ablation) | CGCD framework · CUB-200 (ResNet-18) | M_all: 61.34 -> 45.28 | Hurt |
| run second incremental step (two-step setting) | CGCD framework · CUB-200 two-step | old-class cluster accuracy: decreases by about 20% | Hurt |
Closest papers by meaning
- Solving the Catastrophic Forgetting Problem in Generalized Category Discovery (CVPR 2024 · similarity 0.86)
- Prior-Constrained Association Learning for Fine-Grained Generalized Category Discovery (AAAI 2025 · similarity 0.85)
- Happy: A Debiased Learning Framework for Continual Generalized Category Discovery (NeurIPS 2024 · similarity 0.85)
- A Unified Knowledge Transfer Network for Generalized Category Discovery (AAAI 2024 · similarity 0.85)
- SEAL: Semantic-Aware Hierarchical Learning for Generalized Category Discovery (NeurIPS 2025 · similarity 0.85)
- Continual Generalized Category Discovery: Learning and Forgetting from a Bayesian Perspective (ICML 2025 · 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.
