RAL: Reinforcement active learning for network traffic monitoring and analysis

Abstract

Network-traffic data usually arrives in the form of a data stream. Online monitoring systems need to handle the incoming samples sequentially and quickly. These systems regularly need to get access to ground-truth data to understand the current state of the application they are monitoring, as well as to adapt the monitoring application itself. However, with in-the-wild network-monitoring scenarios, we often face the challenge of limited availability of such data. We introduce RAL, a novel stream-based, active-learning approach, which improves the ground-truth gathering process by dynamically selecting the most beneficial measurements, in particular for model-learning purposes.