Network Traffic Classification 3.0
* Register (or log in) to the AI4G Neural Network to add this session to your agenda or watch the replay
Network operators consistently face challenges posed by the tremendous growth and diversification of traffic in their networks, such as mobile/IoT devices and anonymity tools. Additionally, dynamicity and encryption create further obstacles for actionable network traffic analytics. This talk will cover novel breakthroughs in network Traffic Classification (TC) generated by mobile applications. TC provides valuable network visibility, benefiting network management, user-tailored experiences, and privacy.
In particular, the second wave of data-driven techniques in TC (i.e., the adoption of Deep Learning since 2015, following Machine Learning attempts in the early 2000s) has underscored domain-specific pitfalls and the need for sophisticated architectures to handle the heterogeneous and structured nature of traffic data. However, key shortcomings still hinder the development of ready-to-deploy Deep Learning traffic classifiers. Ideally, these classifiers should be “trustworthy” and “long-lasting,” though such characteristics are rarely found in practice.
Therefore, the innovative application of eXplainable AI (XAI) techniques in designing interpretable and reliable traffic classifiers will be introduced. Additionally, initial efforts to capitalize on class-incremental techniques, enabling adaptation to sudden and inevitable changes in the traffic landscape (e.g., new emerging apps to be managed), will be discussed. The talk will also cover current research in the field of AI-based network traffic analysis conducted by the TRAFFIC group at the University of Naples Federico II, Italy.