Wang, Bo, Junzhuo Li, Hong Chen, Yuanlin Chu, Yuxuan Fan, and Xuming Hu. 2026. “Deconstructing Pre-Training: Knowledge Attribution Analysis in MoE and Dense Models”. Proceedings of the AAAI Conference on Artificial Intelligence 40 (39):33359-67. https://doi.org/10.1609/aaai.v40i39.40622.