Research map / Federated learning and privacy / Personalized and heterogeneous FL
Controllable Federated Prompt Learning at Test Time
Rui Zhu, Liang Bai, Yanming Guo, Yirun Ruan, Tianyuan Yu, Zhihe Lu
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 |
|---|---|---|---|
| use MoDA alignment score instead of Gini index only (ablation) | COTE · CIFAR100 Avg | accuracy: 67.14 -> 71.08 | Improved |
| use MoDA alignment score instead of entropy only (ablation) | COTE · CIFAR100 Avg | accuracy: 70.89 -> 71.08 | Improved |
| use MoDA alignment score instead of regularity index only (ablation) | COTE · CIFAR100 Avg | accuracy: 67.15 -> 71.08 | Improved |
| add confidence-based sample partition (both branches) vs high-confidence branch only (ablation) | COTE · CIFAR100 Avg | accuracy: 65.04 -> 71.08 | Improved |
| add confidence-based sample partition vs low-confidence branch only (ablation) | COTE · ImageNet Avg | accuracy: 75.21 -> 81.44 | Improved |
| add confidence-based partition on ImageNet-R (ablation) | COTE · ImageNet-R | accuracy: 89.37 -> 89.27 | No clear effect |
| increase stability coefficient lambda from 0.7 to 0.9 (ablation) | COTE · CIFAR100 and ImageNet Avg | accuracy: best 71.08 and 81.44 at 0.9 | Improved |
| set phi to extreme values 0 or 1 instead of mid-range (ablation) | MoDA · CIFAR100 and ImageNet | accuracy | Hurt |
| adaptive phi (auto) vs best fixed phi (ablation) | MoDA · CIFAR100 and ImageNet | accuracy | No clear effect |
| personalize prompts via fine-tuning (PromptFL + FT) vs global PromptFL | PromptFL · CIFAR100 OoC | accuracy: 62.30 -> 22.07 | Hurt |
| add generic TTA (TENT) on personalized prompt | PromptFL + FT · CIFAR100 Avg | accuracy: 65.04 -> 51.99 | Hurt |
| use COTE instead of TPT on out-of-client data | personalized FPL · CIFAR100 OoC | accuracy: 19.24 -> 42.34 | Improved |
Closest papers by meaning
- Harmonizing Generalization and Personalization in Federated Prompt Learning (ICML 2024 · similarity 0.92)
- FLiP: Towards Comprehensive and Reliable Evaluation of Federated Prompt Learning (NeurIPS 2025 · similarity 0.91)
- FedPHA: Federated Prompt Learning for Heterogeneous Client Adaptation (ICML 2025 · similarity 0.91)
- Mixture of Experts Made Personalized: Federated Prompt Learning for Vision-Language Models (ICLR 2025 · similarity 0.90)
- Global Prompt Refinement with Non-Interfering Attention Masking for One-Shot Federated Learning (NeurIPS 2025 · similarity 0.90)
- FOCoOp: Enhancing Out-of-Distribution Robustness in Federated Prompt Learning for Vision-Language Models (ICML 2025 · similarity 0.89)
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.
