FedAPT improves adversarial robustness of federated prompt tuning for CLIP by generating visual prompts from text prompts under a global-label beacon, with reported gains of up to 11.49% under PGD-100.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
FedAPT: Federated Adversarial Prompt Tuning for Vision-Language Models
FedAPT improves adversarial robustness of federated prompt tuning for CLIP by generating visual prompts from text prompts under a global-label beacon, with reported gains of up to 11.49% under PGD-100.