Research map / Vision-language and multimodal / CLIP and zero-shot
PSP: Prompt-Guided Self-Training Sampling Policy for Active Prompt Learning
Sen Tao, Kaiduo Feng, Jiawei Liu, Peng Zeng, Yongchao Xu, Yufei Zheng, Zheng-Jun Zha
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 |
|---|---|---|---|
| replace PCB sampling with PSP (VSSP + UST) | PCB (AS) with BADGE · Average over 7 datasets | final accuracy: 74.36 -> 76.87 | Improved |
| PSP vs PCB (AS) on large-scale categories | BADGE + PCB (AS) · Stanford Cars | final accuracy: 70.70 -> 73.84 | Improved |
| PSP vs best baseline on Flowers102 and Caltech101 | BADGE + PCB (AS) · Flowers102 | final accuracy: 96.71 -> 96.35 | No clear effect |
| add vectorized soft actor-critic sampling policy (VSSP) (ablation) | PCB + UST · DTD/Pets/EuroSAT/Aircraft avg | final accuracy: 67.26 -> 68.52 | Improved |
| add uncertainty-augmented self-training (UST) pseudo-labels (ablation) | PCB + VSSP · DTD/Pets/EuroSAT/Aircraft avg | final accuracy: 66.42 -> 68.52 | Improved |
| use state with classification-score gradient embeddings instead of features only (ablation) | VSSP · DTD | final accuracy: 63.36 -> 65.66 | Improved |
| set beta=0.7 vs beta=0.0 in hybrid reward (ablation) | PSP · DTD | final accuracy: 63.89 -> 65.66 | Improved |
| replace UPL* pseudo-labeling with UST (ablation) | prompt learning on CLIP · DTD | final accuracy: 55.08 -> 62.65 | Improved |
| replace XPL with UST (ablation) | prompt learning on CLIP · DTD | final accuracy: 62.29 -> 62.65 | No clear effect |
| swap CLIP for SigLIP, PSP vs PCB | active prompt learning · avg of 4 datasets | final accuracy: 48.82 -> 59.39 | Improved |
| PSP vs CoOp/CoCoOp with same real labels | prompt learning · Average over 7 datasets | final accuracy: 69.11 -> 76.87 (CoOp) | Improved |
Closest papers by meaning
- Black-Box Test-Time Prompt Tuning for Vision-Language Models (AAAI 2025 · similarity 0.88)
- Exploring Task-Level Optimal Prompts for Visual In-Context Learning (AAAI 2025 · similarity 0.87)
- Prompt Tuning Decision Transformers with Structured and Scalable Bandits (NeurIPS 2025 · similarity 0.87)
- Active Prompt Learning in Vision Language Models (CVPR 2024 · similarity 0.87)
- Make Prompts Adaptable: Bayesian Modeling for Vision-Language Prompt Learning with Data-Dependent Prior (AAAI 2024 · similarity 0.86)
- Revisiting the Power of Prompt for Visual Tuning (ICML 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.
