Research map / Continual learning / Class-incremental
Naming to Learn: Class Incremental Learning for Vision-Language Model with Unlabeled Data
Qiwei Li, Xiaochen Yang, Jiahuan Zhou
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 inter-class weight adjustment (ablation) | N2L base (RAIL) · Aircraft (B0Inc10) | final-task accuracy AB (%): 33.59 -> 35.31 | Improved |
| add intra-class weight adjustment (ablation) | N2L base (RAIL) · Aircraft (B0Inc10) | final-task accuracy AB (%): 33.59 -> 34.35 | Improved |
| add pseudo-label refinement (ablation) | N2L base (RAIL) · Aircraft (B0Inc10) | final-task accuracy AB (%): 33.59 -> 35.64 | Improved |
| add pseudo-label refinement to inter+intra adjustment (ablation) | N2L · Aircraft (B0Inc10) | final-task accuracy AB (%): 36.45 -> 40.21 | Improved |
| add intra-class adjustment to inter + refinement (ablation) | N2L · Aircraft (B0Inc10) | final-task accuracy AB (%): 39.66 -> 40.21 | Improved |
| add intra-class adjustment (Gaussian N(1,1/4) weights) (ablation) | N2L · Aircraft (B0Inc10) | final-task accuracy AB (%): 39.69 -> 40.21 | Improved |
| use 1/E entropy weights for intra-class adjustment instead of none (ablation) | N2L · Aircraft (B0Inc10) | final-task accuracy AB (%): 39.69 -> 38.45 | Hurt |
| add CPL pseudo labels | RAIL · Aircraft (B0Inc10) | final-task accuracy AB (%): 33.59 -> 35.80 | Improved |
| add CPL pseudo labels | N2L · Aircraft (B0Inc10) | final-task accuracy AB (%): 40.21 -> 42.99 | Improved |
| add CPL pseudo labels | ENGINE · Aircraft (B0Inc10) | final-task accuracy AB (%): 25.41 -> 25.47 | No clear effect |
| use N2L instead of RAIL with CLIP pseudo labels | class-incremental VLM · Aircraft (B0Inc10) | final-task accuracy AB (%): 33.59 -> 40.21 | Improved |
| train N2L with ground-truth labels instead of pseudo labels | N2L · Aircraft (B0Inc10) | final-task accuracy AB (%): 40.21 -> 56.31 | Improved |
Closest papers by meaning
- Candidate Pseudolabel Learning: Enhancing Vision-Language Models by Prompt Tuning with Unlabeled Data (ICML 2024 · similarity 0.90)
- Generative Multi-modal Models are Good Class Incremental Learners (CVPR 2024 · similarity 0.89)
- Advancing Cross-domain Discriminability in Continual Learning of Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- Rebalancing Multi-Label Class-Incremental Learning (AAAI 2025 · similarity 0.87)
- Handling Imbalanced Pseudolabels for Vision-Language Models with Concept Alignment and Confusion-Aware Calibrated Margin (ICML 2025 · similarity 0.87)
- L3A: Label-Augmented Analytic Adaptation for Multi-Label Class Incremental Learning (ICML 2025 · similarity 0.87)
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.
