Research map / Federated learning and privacy / Personalized and heterogeneous FL
Latte: Collaborative Test-Time Adaptation of Vision-Language Models in Federated Learning
Wenxuan Bao, Ruxi Deng, Ruizhong Qiu, Tianxin Wei, Hanghang Tong, Jingrui He
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 |
|---|---|---|---|
| share one global memory across all clients | TDA · VLCS (ViT-B/16) | accuracy (%): 81.44 -> 80.29 | Hurt |
| share one global memory across all clients | DMN-ZS · VLCS (ViT-B/16) | accuracy (%): 81.12 -> 80.55 | Hurt |
| share one global memory across all clients | DMN-ZS · CIFAR-10-C (ViT-B/16) | accuracy (%): 67.42 -> 63.90 | Hurt |
| share one global memory across all clients | DMN-ZS · TerraIncognita (ViT-B/16) | accuracy (%): 33.65 -> 37.64 | Improved |
| share one global memory across all clients | TDA · TerraIncognita (RN50) | accuracy (%): 36.73 -> 37.78 | Improved |
| use Latte local plus external memory instead of global sharing | DMN-ZS (global) · TerraIncognita (ViT-B/16) | accuracy (%): 37.64 -> 40.95 | Improved |
| use Latte instead of best baseline Zero | Zero · VLCS (ViT-B/16) | accuracy (%): 81.98 -> 82.57 | Improved |
| use Latte instead of CLIP zero-shot | CLIP · TerraIncognita (RN50) | accuracy (%): 23.24 -> 41.47 | Improved |
| use Latte instead of CLIP zero-shot | CLIP · CIFAR-100-C (ViT-B/16) | accuracy (%): 38.91 -> 39.51 | Improved |
| add memory-based TTA on CIFAR-100-C vs episodic TPT/Zero | Zero · CIFAR-100-C (ViT-B/16) | accuracy (%): 38.91 -> 37.55 | Hurt |
| update external memory every T samples up to 50 (ablation) | Latte · TerraIncognita | accuracy (%): T<=50 unchanged; T = 200 slight drop | No clear effect |
| add external memory to local memory (ablation) | Latte (local only) · VLCS | accuracy (%): local only or external only -> merged (no numbers given) | Improved |
Closest papers by meaning
- DOTA: Distributional Test-time Adaptation of Vision-Language Models (NeurIPS 2025 · similarity 0.90)
- Efficient Test-Time Adaptation of Vision-Language Models (CVPR 2024 · similarity 0.88)
- Efficient Test-time Adaptation through Candidate Verification and Divergence Shifts (NeurIPS 2026 · similarity 0.87)
- Frustratingly Easy Test-Time Adaptation of Vision-Language Models (NeurIPS 2024 · similarity 0.87)
- BoostAdapter: Improving Vision-Language Test-Time Adaptation via Regional Bootstrapping (NeurIPS 2024 · similarity 0.87)
- SCAP: Transductive Test-Time Adaptation via Supportive Clique-based Attribute Prompting (CVPR 2025 · similarity 0.87)
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.
