Research map / Vision-language and multimodal / CLIP and zero-shot
Enhancing Continual Learning of Vision-Language Models via Dynamic Prefix Weighting
Hyeonseo Jang, Hyuk Kwon, Kibok Lee
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 RePA to prefix tuning (ablation) | DPW baseline · MTIL Order I | Last: 85.9 -> 86.4 | Improved |
| add CondAct (replace softmax) (ablation) | DPW baseline · MTIL Order I | Avg.: 76.6 -> 77.8 | Improved |
| add RePA on top of CondAct (ablation) | DPW · MTIL Order I | Avg.: 77.8 -> 78.9 | Improved |
| add RWM adapter weighting (ablation) | DPW · MTIL Order I | Last: 87.9 -> 88.3 | Improved |
| apply RePA to input tokens instead of retrained QK projection (ablation) | RePA · MTIL Order I | Avg.: 76.8 -> 79.3 | Improved |
| apply RePA to input tokens rather than on top of query projection (ablation) | RePA · MTIL Order I | Avg.: 78.6 -> 79.3 | Improved |
| add sigmoid activation to RePA (ablation) | CondAct · MTIL Order I | Avg.: 76.8 -> 78.3 | Improved |
| add conditional normalization after sigmoid (ablation) | CondAct · MTIL Order I | Transfer: 68.6 -> 69.9 | Improved |
| add conditional filtering (ablation) | CondAct · MTIL Order I | Transfer: 69.9 -> 70.4 | Improved |
| double prefix length (ablation) | DPW without RWM · MTIL Order I | Avg.: 78.9 -> 78.4 | Hurt |
| use learnable router for adapter instead of RWM (ablation) | DPW · MTIL Order I | Avg.: 78.9 -> 79.0 | No clear effect |
| freeze shared down projection D (ablation) | adapter · MTIL Order I | Last: 25.7 -> 88.0 | Improved |
Closest papers by meaning
- Synthetic Data is an Elegant GIFT for Continual Vision-Language Models (CVPR 2025 · similarity 0.89)
- Boosting Continual Learning of Vision-Language Models via Mixture-of-Experts Adapters (CVPR 2024 · similarity 0.89)
- Advancing Cross-domain Discriminability in Continual Learning of Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- Frustratingly Easy Test-Time Adaptation of Vision-Language Models (NeurIPS 2024 · similarity 0.87)
- Towards Self-Refinement of Vision-Language Models with Triangular Consistency (NeurIPS 2025 · similarity 0.87)
- Skip Tuning: Pre-trained Vision-Language Models are Effective and Efficient Adapters Themselves (CVPR 2025 · similarity 0.86)
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.
