V-SEAM: Visual Semantic Editing and Attention Modulating for Causal Interpretability of Vision-Language Models.

Qidong Wang, Junjie Hu, Ming Jiang

Browse the full EMNLP paper archive.