Research map / Vision-language and multimodal / CLIP and zero-shot
DVLA-RL: Dual-Level Vision–Language Alignment with Reinforcement Learning Gating for Few-Shot Learning
Wenhao Li, Xianjing Meng, Qiangchang Wang, Zhongyi Han, Zhibin Wu, Yilong Yin
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 attribute-level semantics and RL-gated attention to the baseline (ablation) | baseline · miniImageNet 1-shot | accuracy: 68.67 -> 75.73 | Improved |
| add description-level semantics to Attr + Top-k + RLA (ablation) | DVLA-RL · miniImageNet 1-shot | accuracy: 76.56 -> 81.69 | Improved |
| add progressive Top-k attribute selection (ablation) | DVLA-RL Attr + RLA · miniImageNet 1-shot | accuracy: 75.73 -> 76.56 | Improved |
| add progressive Top-k attribute selection (ablation) | DVLA-RL Attr + RLA · miniImageNet 5-shot | accuracy: 85.76 -> 85.25 | Hurt |
| add progressive Top-k attribute selection (ablation) | DVLA-RL Attr + RLA · CUB 1-shot | accuracy: 88.63 -> 89.33 | Improved |
| add description-level semantics to attribute-level model (ablation) | DVLA-RL Attr + Top-k + RLA · CUB 5-shot | accuracy: 90.91 -> 95.06 | Improved |
| replace summation gating with RL gate (ablation) | DVLA-RL · miniImageNet 1-shot | accuracy: 79.63 -> 81.69 | Improved |
| replace MLP gating with RL gate (ablation) | DVLA-RL · miniImageNet 1-shot | accuracy: 80.05 -> 81.69 | Improved |
| replace ECER fusion with RLA module (ablation) | DVLA-RL · miniImageNet 1-shot | accuracy: 79.35 -> 81.69 | Improved |
| use DSC semantics instead of class-name semantics (ablation) | DVLA-RL · miniImageNet 1-shot | accuracy: 78.96 -> 81.69 | Improved |
| increase Beta concentration kappa from 5 to 20 (ablation) | DVLA-RL · miniImageNet 1-shot | accuracy: 81.69 -> 81.58 | No clear effect |
| increase RL weight lambda from 0.1 to 1.0 (ablation) | DVLA-RL · miniImageNet 1-shot | accuracy: 81.69 -> 81.08 | Hurt |
Closest papers by meaning
- VT-FSL: Bridging Vision and Text with LLMs for Few-Shot Learning (NeurIPS 2025 · similarity 0.88)
- VL-SAE: Interpreting and Enhancing Vision-Language Alignment with a Unified Concept Set (NeurIPS 2025 · similarity 0.88)
- CogVLA: Cognition-Aligned Vision-Language-Action Models via Instruction-Driven Routing & Sparsification (NeurIPS 2025 · similarity 0.88)
- Fine-Grained Verifiers: Preference Modeling as Next-token Prediction in Vision-Language Alignment (ICLR 2025 · similarity 0.87)
- What Can RL Bring to VLA Generalization? An Empirical Study (NeurIPS 2025 · similarity 0.87)
- MMA: Multi-Modal Adapter for Vision-Language Models (CVPR 2024 · 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.
