On Explaining and Reasoning About Optical Fiber Link Problems

Optical fiber links are known for their high bandwidth and reliable data transmission. However, problems may still arise, affecting signal quality and network performance. These problems are usually happening due to external physical extrusion or excessive bending, insufficient transmission power, damaged connectors causing signal loss; or failures of splice tray connector. In response to increasing optical fiber link problems transparency and interpetability, various attempts have been made to bring explainability in Artificial Intelligence (AI) decision-making and reasoning processes. This paper tackles a crucial and timely topic, i.e., understand the various factors contributing to optical link problems by explaining opaque AI models with two goals: (i) either providing instance explanations for a given decision by using a local and model agnostic approach; or (ii) providing global explanations able to describe the overall logic assuming knowledge of the black box model or its internals. The scientific contribution of this paper entails novel explainable AI (XAI) models harvesting data from optical fiber link events to first derive local explanations, and then apply a hierarchical approach to educe global explanations from the local ones. The proposed approach shows that we can efficiently tackle both explanation complexity and fidelity to reason about the causes that have resulted in optical fiber link problems.