“Mousika: Enable General In-Network Intelligence in Programmable Switches by Knowledge Distillation” is a research paper authored by Guorui Xie, Qing Li, Yutao Dong, Guang-Hua Duan, Yong Jiang, and Jingpu Duan. The paper addresses the challenges of implementing complex learning models, such as deep learning, on programmable switches due to their high computational complexity and large storage requirements. To overcome these limitations, the authors propose Mousika, an in-network intelligence framework that utilizes Binary Decision Trees (BDT) instead of traditional Decision Trees (DT) for faster training, fewer rules, and better compatibility with switch constraints. Additionally, Mousika introduces a teacher-student knowledge distillation process that allows the translation of other learning models to BDT, enabling the deployment of advanced learning capabilities on switches without direct computational and memory constraints. This framework was presented at the IEEE INFOCOM 2022 - IEEE Conference on Computer Communications.
Reference List
- Xie, G., Li, Q., Dong, Y., Duan, G., Jiang, Y., & Duan, J. (2022, May). Mousika: Enable general in-network intelligence in programmable switches by knowledge distillation. In IEEE INFOCOM 2022-IEEE Conference on Computer Communications (pp. 1938-1947). IEEE.