Enabling Self-Driving Networks with Machine Learning

Abstract

This work aims to enable self-driving networks by addressing the lack of trust that network operators have in Machine Learning (ML) models for networking problems. To achieve this, we propose a natural-language conversational interface (LUMI) and a new ML pipeline that uses techniques from the emerging field of eXplainable Artificial Intelligence (XAI) to scrutinize models. We also propose a new XAI method called TRUSTEE to extract explanations from any given black-box ML model in the form of decision trees of manageable sizes. Our results indicate that ML models used in networking problems need to be put under proper scrutiny and corrected to fulfill their tasks properly.