The author explores the impact of feature shift on generalization in vision-language prompt learning, proposing RESTORE to address misalignment and improve performance.
