Abstract:Network traffic classification aims to associate the network traffic with a class of traffic characterization (e.g., Streaming) or applications (e.g., Facebook). This ability plays an important role in advanced network management. The tasks of traffic characterization and application identification are usually implemented by individual models. However, when multiple models are deployed in the online environment, this causes a dramatic increase in the complexity, resource demand and maintenance costs. In this paper, an efficient multi-task learning method named multi-task transformer (MTT) is proposed. It simultaneously classifies the traffic characterization and application identification tasks. The proposed model considers the input packet as a sequence of bytes and applies a multi-head attention mechanism to extract features. Experiments are conducted on the ISCX VPN-nonVPN dataset to demonstrate the effectiveness of MTT. F1\documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$F_1$$\end{document} scores of 98.75% and 99.35% have been achieved for application identification and traffic characterization, respectively. To the best of our knowledge, the results are better than the state-of-the-art results. The MTT model outputs the two results simultaneously in ∼\documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$\sim$$\end{document} 0.1 milliseconds (per packet), which satisfies the requirement of online traffic classification. Compared with the 1D-CNN and 2D-CNN models, the proposed MTT model is more stable, presents higher classification performance and requires less storage space. Finally, the selection strategies of input length for different neural networks are suggested and the related principles are investigated.

A Multi-Modal Approach For Context-Aware Network Traffic Classification

A Cost-Sensitive Deep Learning-Based Approach for Network Traffic Classification

Network Traffic Classification Via Non-Convex Multi-Task Feature Learning

Novel Dynamic Multiple Classification System for Network Traffic

MATEC: A lightweight neural network for online encrypted traffic classification

Multi-model induced network for participatory-sensing-based classification tasks in intelligent and connected transportation systems.

MTC-NET: A Multi-Channel Independent Anomaly Detection Method for Network Traffic

Traffic Flow and Speed Forecasting Through a Bayesian Deep Multi-Linear Relationship Network.

L2-BiTCN-CNN: Spatio-temporal features fusion-based multi-classification model for various internet applications identification

TCGNN: Packet-grained Network Traffic Classification Via Graph Neural Networks

Multitask Learning for Network Traffic Classification

Res2Net-based multi-scale and multi-attention model for traffic scene image classification

A unified multimodal classification framework based on deep metric learning

Multi-scale Convolutional Feature Fusion Network Based on Attention Mechanism for IoT Traffic Classification

A Multimodal Network Security Framework for Healthcare Based on Deep Learning

MTT: an efficient model for encrypted network traffic classification using multi-task transformer

A Novel Self-Supervised Framework Based on Masked Autoencoder for Traffic Classification

Network Traffic Classification Model Based on Spatio-Temporal Feature Extraction

Method for multi-task learning fusion network traffic classification to address small sample labels

Network traffic grant classification based on 1DCNN-TCN-GRU hybrid model

A New Network Traffic Classification Method Based on Classifier Integration