미지 트래픽 분류 연구 동향

초록

This paper surveys unknown traffic classification techniques for an open-world scenario. We review coarse-grained methods (single unknown class) and fine-grained methods based on representation learning. While state-of-the-art models classify known traffic well, they show low performance in fine-grained unknown classification. We find that this limitation stems from fixed decision boundaries formed by simple clustering. We propose that future research must focus on dynamic decision boundary formation.