Many or few samples? Comparing transfer, contrastive and meta-learning in encrypted traffic classification

Guarino, Idio; Wang, Chao; Finamore, Alessandro; Pescape, Antonio; Rossi, Dario
TMA 2023, 7th IFIP Network Traffic Measurement and Analysis Conference, 26-29 June 2023, Naples, Italy

The popularity of Deep Learning (DL), coupled with network traffic visibility reduction due to the increased adoption of HTTPS, QUIC, and DNS-SEC, re-ignited interest towards Traffic Classification (TC). However, to tame the dependency from task-specific large labeled datasets, we need to find better ways to learn representations that are valid across tasks. In this work we investigate this problem comparing transfer learning,
meta-learning and contrastive learning against reference Machine Learning (ML) tree-based and monolithic DL models (16 methods total). Using two publicly available datasets, namely MIRAGE19 (40 classes) and AppClassNet (500 classes), we show that (i) by using DL methods on large datasets we can obtain more general representations with (ii) contrastive learning methods yielding the best performance and (iii) meta-learning the worst one. While (iv) tree-based models can be impractical for large
tasks but fit well small tasks, (v) DL methods that reuse better learned representations are closing their performance gap against trees also for small tasks.

DOI
Type:
Conférence
City:
Naples
Date:
2023-06-26
Department:
Data Science
Eurecom Ref:
7585
Copyright:
© IFIP. Personal use of this material is permitted. The definitive version of this paper was published in TMA 2023, 7th IFIP Network Traffic Measurement and Analysis Conference, 26-29 June 2023, Naples, Italy and is available at : http://dx.doi.org/10.23919/TMA58422.2023.10198965

PERMALINK : https://www.eurecom.fr/publication/7585