Research map / Vision-language and multimodal / CLIP and zero-shot
BioMedVR: Confusion-Aware Mixture-of-Prompt Experts for Biomedical Visual Reprogramming
Jiaxiang Liu, Tianxiang Hu, Juwei Guan, Yujie Wu, Yusong Wang, Yao Mu, Zuozhu Liu, Mingkun Xu
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 confusion-aware attributes (negative expert text) (ablation) | BioMedVR · 11 biomedical datasets avg | accuracy (%): 71.4 -> 76.4 | Improved |
| add confusion-suppression loss (ablation) | BioMedVR · 11 biomedical datasets avg | accuracy (%): 73.6 -> 76.4 | Improved |
| add negative prompt expert (MoPE) (ablation) | BioMedVR · 11 biomedical datasets avg | accuracy (%): 75.6 -> 76.4 | Improved |
| add negative prompt expert (MoPE) (ablation) | BioMedVR · COVID-19 | accuracy (%): 74.3 -> 77.4 | Improved |
| use confusion-aware VR instead of AttrVR | AttrVR · 11 biomedical datasets avg | accuracy (%): 73.0 -> 76.4 | Improved |
| use GPT-4.1 instead of GPT-5-mini/GPT-5 for attribute generation (ablation) | BioMedVR · 11 biomedical datasets avg | accuracy (%): 76.4 -> 75.2 | No clear effect |
| use GPT-4o-mini instead of GPT-4.1 for attribute generation (ablation) | BioMedVR · 11 biomedical datasets avg | accuracy (%): 76.4 -> 76.4 | No clear effect |
| apply BioMedVR to RN50 backbone | AttrVR baseline · 11 biomedical datasets avg (RN50) | accuracy (%): 49.0 -> 51.6 | Improved |
| apply BioMedVR to ViT-B/32 backbone | AttrVR baseline · 11 biomedical datasets avg (ViT-B/32) | accuracy (%): 59.0 -> 60.5 | Improved |
| add confusion-aware zero-shot scoring | CLIP · 11 biomedical datasets avg (zero-shot) | accuracy (%): 28.1 -> 31.6 | Improved |
| add confusion-aware zero-shot scoring | CLIP · EuroSAT (zero-shot) | accuracy (%): 21.2 -> 21.0 | No clear effect |
| doubling VR parameters with two experts | AttrVR · Training cost (BUSI) | trainable parameters: 0.15M -> 0.30M | Hurt |
Closest papers by meaning
- BiomedCoOp: Learning to Prompt for Biomedical Vision-Language Models (CVPR 2025 · similarity 0.89)
- LLaDA-MedV: Exploring Large Language Diffusion Models for Biomedical Image Understanding (CVPR 2026 · similarity 0.88)
- Sample-specific Masks for Visual Reprogramming-based Prompting (ICML 2024 · similarity 0.88)
- Understanding Model Reprogramming for CLIP via Decoupling Visual Prompts (ICML 2025 · similarity 0.87)
- LLMind: Bio-inspired Training-free Adaptive Visual Representations for Vision-Language Models (CVPR 2026 · similarity 0.87)
- pMoE: Prompting Diverse Experts Together Wins More in Visual Adaptation (ICLR 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.
