Xu, R.; Wang, Y.; Luo, Y.; Du, B. Rethinking Visual Token Reduction in LVLMs Under Cross-Modal Misalignment. AAAI 2026, 40, 27323-27331.