Prompt-Anchored Residual Adaptation for Biomedical Vision-Language Models
PARA is proposed, which retains the frozen prompt prediction as a support-invariant semantic anchor and incorporates a visual prediction learned from the support set through an anchor-relative residual and achieves state-of-the-art performance in both few-shot classification and base-to-novel generalization.
Jing-Xuan Kang, Qian-Ying Yue, Che Liu et al.
· 0 citations