Abstract This paper introduces a novel approach to incorporate XAI in adaptive optical transport systems that aims to increase the interpretability and transparency of AI-based decisions. The research focuses on two key goals: (1) to develop and test how to design interpretable AI models, which can explain their decisions in optical network management tasks, and (2) to measure the effect of explainability on operator trust, fault localization, and overall system performance. The proposed architecture employs well-known XAI methods (e.g., SHAP, LIME, attention-based visualization) combined with supervised and reinforcement learning, for functions’ dynamic bandwidth allocation, path reconfiguration, and for the root-cause failure analysis.
Kumar et al. (Thu,) studied this question.