Bootstrapping Trust in ML4Nets Solutions with Hybrid Explainability
Abduarraheem Elfandi, Hannah Sagalyn, Ramakrishnan Durairajan, Walter Willinger · 2024
The future success of ML4Nets---defined as the application of machine learning (ML) techniques to address real-world network security and performance problems---relies critically on convincing network operators to deploy ML-based solutions in their production networks. However, the black-box nature of many of these solutions has been a major impediment to both gaining the operators' trust in the underlying trained models and providing effective safety guarantees. Explainable AI (XAI) represents an approach to dealing with this problem and provides operators with global and local explainability techniques that enable them to reason about the decisions made by trained ML models. Unfortunately, these solutions are lacking in simultaneously engendering the kind of trust and ensuring the type of safety guarantees that operators require for using ML solutions in practice.