Research map / Vision-language and multimodal / CLIP and zero-shot
Verbalized Representation Learning for Interpretable Few-Shot Generalization
Cheng-Fu Yang, Da Yin, Wenbo Hu, Heng Ji, Nanyun Peng, Bolei Zhou, Kai-Wei Chang
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 commonality features F_comm to difference features F_diff (ablation) | VRL · iNaturalist feature fusion (avg) | accuracy: 65.26 -> 67.92 | Improved |
| add CLIP image features to F_diff and F_comm (ablation) | VRL (concat) · iNaturalist feature fusion (avg) | accuracy: 67.92 -> 73.92 | Improved |
| ensemble per-feature classifiers instead of concatenation (ablation) | VRL · iNaturalist feature fusion (avg) | accuracy: 73.92 -> 76.52 | Improved |
| soft voting instead of hard voting (LR, LR, MLP) (ablation) | VRL ensemble · iNaturalist feature fusion (avg) | accuracy: 71.60 -> 76.52 | Improved |
| add DINO features to F_diff, F_comm and CLIP (ablation) | VRL ensemble · iNaturalist feature fusion (avg) | accuracy: 76.52 -> 79.92 | Improved |
| use VRL instead of LLaVA SFT | LLaVA · iNaturalist (10 images per class) | accuracy: 43.9 -> 62.1 | Improved |
| use VRL instead of LLaVA SFT | LLaVA · iNaturalist (200 images per class) | accuracy: 80.2 -> 86.2 | Improved |
| use VRL instead of in-context learning | LLaVA-7B · iNaturalist (5 families) | accuracy: 22.92 -> 58.46 | Improved |
| use VRL-extracted features instead of human-labeled attributes | VRL · Kiki-Bouba (human attributes comparison) | accuracy: 63.15 -> 83.00 | Improved |
| scale VRL from 7B to 72B (ablation) | LLaVA-VRL · iNaturalist (5 families) | accuracy: 58.46 -> 67.92 | Improved |
| scale VRL from 7B to 72B (ablation) | LLaVA-VRL · Kiki-Bouba (avg of 2 splits) | accuracy: 83.00 -> 81.90 | No clear effect |
| use LLaVA instead of CLIP (binary) for feature mapping with LR (ablation) | VRL · iNaturalist | accuracy: 58.0 -> 67.4 | Improved |
Closest papers by meaning
- Black Sheep in the Herd: Playing with Spuriously Correlated Attributes for Vision-Language Recognition (ICLR 2025 · similarity 0.87)
- MMRL: Multi-Modal Representation Learning for Vision-Language Models (CVPR 2025 · similarity 0.87)
- Few-Shot Recognition via Stage-Wise Retrieval-Augmented Finetuning (CVPR 2025 · similarity 0.87)
- Making Large Vision Language Models to Be Good Few-Shot Learners (AAAI 2025 · similarity 0.87)
- Leveraging Vision-Language Models for Improving Domain Generalization in Image Classification (CVPR 2024 · similarity 0.87)
- Revisiting Few-Shot Object Detection with Vision-Language Models (NeurIPS 2024 · 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.
