Research map / Label-efficient and robust learning / Test-time adaptation
USE: A Unified Self-Ensembling Framework for Test-Time Prompt Tuning
Siru Jiang, Jian Liang, Ran He, Tieniu Tan
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 |
|---|---|---|---|
| apply SE ensembling at inference instead of weak view only (ablation) | TPT-style RCE tuning · ViT-B/16 avg | accuracy: 63.71 -> 64.77 | Improved |
| use SE at inference instead of uniform averaging (ablation) | RCE-tuned prompt · ViT-B/16 avg | accuracy: 64.56 -> 64.77 | Improved |
| replace marginal entropy with RCE on SE pseudo label (with SE inference) (ablation) | TPT · ViT-B/16 avg | accuracy: 64.43 -> 64.77 | Improved |
| RCE alone vs marginal entropy with standard inference (ablation) | TPT · ViT-B/16 avg | accuracy: 63.60 -> 63.71 | No clear effect |
| add SE plug-in to C-TPT | C-TPT · ImageNet + OOD (IN&O) | accuracy: 61.09 -> 64.27 | Improved |
| add SE plug-in to TPT | TPT · ViT-B/16 avg | accuracy: 63.60 -> 64.43 | Improved |
| add SE plug-in to STS | STS · ViT-B/16 IN&O | accuracy: 63.89 -> 63.82 | No clear effect |
| SE as optimization-free method vs zero-shot CLIP | CLIP · EuroSAT ViT-B/16 | accuracy: 42.05 -> 44.44 | Improved |
| ZERO marginalization vs zero-shot CLIP | CLIP · EuroSAT ViT-B/16 | accuracy: 42.05 -> 37.19 | Hurt |
| use square-root rescaling of beta instead of linear (ablation) | SE · ViT-B/16 avg | accuracy: 64.34 -> 64.19 | No clear effect |
| add SE with text prompt ensembles | CLIP · ViT-B/16 avg | accuracy: 62.87 -> 65.61 | Improved |
| add skip technique when weak and selected views agree (ablation) | USE · efficiency | accuracy | No clear effect |
Closest papers by meaning
- Black-Box Test-Time Prompt Tuning for Vision-Language Models (AAAI 2025 · similarity 0.92)
- C-TPT: Calibrated Test-Time Prompt Tuning for Vision-Language Models via Text Feature Dispersion (ICLR 2024 · similarity 0.91)
- The Illusion of Progress? A Critical Look at Test-Time Adaptation for Vision-Language Models (NeurIPS 2025 · similarity 0.91)
- Robust Test-Time Adaptation for Zero-Shot Prompt Tuning (AAAI 2024 · similarity 0.90)
- Efficient Test-Time Adaptation of Vision-Language Models (CVPR 2024 · similarity 0.90)
- DynaPrompt: Dynamic Test-Time Prompt Tuning (ICLR 2025 · similarity 0.90)
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.
