<ul><li>A new method called Prompt Gradient Alignment (PGA) is proposed for improving Unsupervised Domain Adaptation (UDA) in vision-language models.</li><li>PGA leverages large-scale pre-trained vision-language models to learn both domain-invariant and specific features.</li><li>The method aligns per-objective gradients to foster consensus between them, and prevents overfitting by penalizing the norm of the gradients.</li><li>Experimental results show that PGA outperforms other vision-language model adaptation methods for UDA.</li></ul>

Enhancing Domain Adaptation through Prompt Gradient Alignment

Discover more