netFound: Principled Design for Network Foundation Models
📰 ArXiv cs.AI
arXiv:2310.17025v5 Announce Type: replace-cross Abstract: Network foundation models promise reusable representations for diverse traffic analysis tasks, but recent diagnostic works have revealed fundamental problems: models exploit dataset shortcuts rather than learning genuine traffic patterns, produce collapsed embedding spaces, and fail to capture the exogenous network conditions that shape real-world behavior. We translate these diagnostic insights into four concrete design principles: proto
DeepCamp AI