Research map / Architectures and efficiency / Neural architecture search
Neural Prompt Search
Yuanhan Zhang, Kaiyang Zhou, Ziwei Liu
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 |
|---|---|---|---|
| search jointly over Adapter, LoRA and VPT design instead of using LoRA alone | LoRA · VTAB-1k average | top-1 accuracy: 74.5 -> 75.5 | Improved |
| search combined prompt design instead of VPT alone | VPT · VTAB-1k average | top-1 accuracy: 72.0 -> 75.5 | Improved |
| search combined prompt design instead of Adapter alone | Adapter · VTAB-1k average | top-1 accuracy: 73.9 -> 75.5 | Improved |
| search design instead of full fine-tuning | Full · VTAB-1k average | top-1 accuracy: 68.9 -> 75.5 | Improved |
| search design instead of LoRA on Specialized group | LoRA · VTAB-1k Specialized | top-1 accuracy: 84.6 -> 84.9 | No clear effect |
| search design instead of LoRA on dSpr-Loc | LoRA · VTAB-1k dSpr-Loc | top-1 accuracy: 75.7 -> 81.8 | Improved |
| search design instead of LoRA on EuroSAT | LoRA · VTAB-1k EuroSAT | top-1 accuracy: 95.3 -> 95.4 | No clear effect |
| use NOAH instead of LoRA for domain generalization | LoRA · ImageNet-V2 | top-1 accuracy: 59.3 -> 66.1 | Improved |
| use NOAH instead of LoRA for domain generalization | LoRA · ImageNet-A | top-1 accuracy: 6.9 -> 11.9 | Improved |
| use NOAH instead of LoRA on source ImageNet | LoRA · ImageNet (16-shot) | top-1 accuracy: 70.8 -> 71.5 | Improved |
| retrain extracted subnet instead of inheriting supernet weights (ablation) | NOAH (inherited) · VTAB-1k average | top-1 accuracy: 74.1 -> 75.5 | Improved |
| retrain subnet on Structured group (ablation) | NOAH (inherited) · VTAB-1k Structured | top-1 accuracy: 58.7 -> 61.3 | Improved |
Closest papers by meaning
- Learning to Learn Better Visual Prompts (AAAI 2024 · similarity 0.91)
- Black-Box Test-Time Prompt Tuning for Vision-Language Models (AAAI 2025 · similarity 0.89)
- Facing the Elephant in the Room: Visual Prompt Tuning or Full finetuning? (ICLR 2024 · similarity 0.89)
- LION: Implicit Vision Prompt Tuning (AAAI 2024 · similarity 0.89)
- Revisiting the Power of Prompt for Visual Tuning (ICML 2024 · similarity 0.88)
- IPO: Interpretable Prompt Optimization for Vision-Language Models (NeurIPS 2024 · similarity 0.88)
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.
