一作ACL 2026Findings· Findings of ACL
DANCE : Diversity-attended Dynamic Caching with Asymmetric Quantization for Test-time Adaptation of Vision-Language Models
Shunge Zou · Changhu Wang · Wei Ju · Ziyue Qiao · Xiao Luo
摘要
This paper studies the problem of test-time adaptation for vision-language models (VLMs). Recent approaches typically measure the prediction entropy to store a confident cache for logit refinement. However, these confident samples tend to approach prototypes with limited coverage of data distribu...