Revealing Epistemic Uncertainty in MLLMs via Causal-Invariant Masking

作者：Haoyang Luo, Linwei Tao, Jie Gui, Xinghao Chen, Chang Xu, Jianyuan Guo, Minjing Dong · 单位：City University of Hong Kong, Southeast University, Huawei Noah’s Ark Lab, University of Sydney · 会议/期刊：arXiv preprint · 方向：cs.CV · 发布日期：2026-10-05