Y. Han, “Filter, Correlate, Compress: Training-Free Token Reduction for MLLM Acceleration”, AAAI, vol. 40, no. 6, pp. 4601–4609, Mar. 2026.