Improving Generalizability of Prompt Learning for Vision-Language Models via Meta-Regularization

ProMetaR, a framework that jointly meta-learns the regularizer and soft prompts, improves the generalizability of prompt tuning for vision-language models.