R. Xu, Y. Wang, Y. Luo, and B. Du, “Rethinking Visual Token Reduction in LVLMs Under Cross-Modal Misalignment”, AAAI, vol. 40, no. 32, pp. 27323–27331, Mar. 2026.