Research map / Label-efficient and robust learning / Semi-supervised learning
Debiased Textual Prompt Tuning for Enhancing Unknown Class Discovery
Yuxin Fan, Junbiao Cui, Changhao Liu, Xingwang Zhao, Jiye Liang
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 textual prompt optimization (optimal-transport debiasing) | TPOBGE · Hybrid | unknown samples misclassified as known: 2.9k -> 1.5k | Improved |
| add textual prompt optimization (optimal-transport debiasing) | TPOBGE · ImageNet-1K | unknown samples misclassified as known: 15k -> 1.6k | Improved |
| add textual prompt optimization (ablation) | TPOBGE · Hybrid, ImageNet-500, ImageNet-1K, Webvision | clustering accuracy | Improved |
| add bimodal graph enhancement on top of prompt optimization (ablation) | TPOBGE · Hybrid, ImageNet-500, ImageNet-1K, Webvision | clustering accuracy | Improved |
| build bimodal proximity graph (more edges) (ablation) | TPOBGE · Hybrid | edge count: 15.1k -> 114k | Improved |
| build bimodal proximity graph (more edges) (ablation) | TPOBGE · ImageNet-1K | edge count: 50k -> 375k | Improved |
| vary tau_T, tau_I, k, beta one at a time (ablation) | TPOBGE · Hybrid | clustering accuracy | No clear effect |
| replace CSC-OWSSL training with frozen encoders and prompt-only tuning | TPOBGE · Hybrid | training time: 543 minutes -> 3 minutes | Improved |
| replace CSC-OWSSL training with frozen encoders and prompt-only tuning | TPOBGE · ImageNet-1K | training time: 7303 minutes -> 10 minutes | Improved |
| TPOBGE versus SpectralGCD (best prior unknown-class average) | TPOBGE · Average over 4 datasets | unknown-class accuracy: 54.03 -> 68.65 | Improved |
| TPOBGE versus SpectralGCD on Hybrid | TPOBGE · Hybrid | all-class accuracy: 63.79 -> 63.81 | No clear effect |
Closest papers by meaning
- Learning Textual Prompts for Open-World Semi-Supervised Learning (CVPR 2025 · similarity 0.91)
- OwMatch: Conditional Self-Labeling with Consistency for Open-World Semi-Supervised Learning (NeurIPS 2024 · similarity 0.87)
- One Prompt Fits All: Universal Graph Adaptation for Pretrained Models (NeurIPS 2025 · similarity 0.86)
- Liberating Seen Classes: Boosting Few-Shot and Zero-Shot Text Classification via Anchor Generation and Classification Reframing (AAAI 2024 · similarity 0.86)
- CGMatch: A Different Perspective of Semi-supervised Learning (CVPR 2025 · similarity 0.86)
- Unknown Prompt the only Lacuna: Unveiling CLIP's Potential for Open Domain Generalization (CVPR 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.
