@inproceedings{su-etal-2025-cartesianmoe,
    title = "{C}artesian{M}o{E}: Boosting Knowledge Sharing among Experts via {C}artesian Product Routing in Mixture-of-Experts",
    author = "Su, Zhenpeng  and
      W, Xing  and
      Lin, Zijia  and
      Xiong, Yizhe  and
      Lv, Minxuan  and
      Ma, Guangyuan  and
      Chen, Hui  and
      Hu, Songlin  and
      Ding, Guiguang",
    editor = "Chiruzzo, Luis  and
      Ritter, Alan  and
      Wang, Lu",
    booktitle = "Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)",
    month = apr,
    year = "2025",
    address = "Albuquerque, New Mexico",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/fix-sig-urls/2025.naacl-long.505/",
    pages = "10040--10055",
    ISBN = "979-8-89176-189-6",
    abstract = "Large language models (LLM) have been attracting much attention from the community recently, due to their remarkable performance in all kinds of downstream tasks. According to the well-known scaling law, scaling up a dense LLM enhances its capabilities, but also significantly increases the computational complexity. Mixture-of-Experts (MoE) models address that by allowing the model size to grow without substantially raising training or inference costs. Yet MoE models face challenges regarding knowledge sharing among experts, making their performance somehow sensitive to routing accuracy. To tackle that, previous works introduced shared experts and combined their outputs with those of the top $K$ routed experts in an addition manner. In this paper, inspired by collective matrix factorization to learn shared knowledge among data, we propose CartesianMoE, which implements more effective knowledge sharing among experts in more like a multiplication manner. Extensive experimental results indicate that CartesianMoE outperforms previous MoE models for building LLMs, in terms of both perplexity and downstream task performance. And we also find that CartesianMoE achieves better expert routing robustness."
}