When Does Visual Prompting Outperform Linear Probing for Vision-Language Models? A Likelihood PerspectiveHsi-Ai TsaoLei Hsiunget al.2025ICASSP 2025Conference paper