Explainability-as-a-Service for Edge AI Systems

A new distributed architecture called Explainability-as-a-Service (XaaS) decouples inference from explanation generation in edge AI systems. Proposed in arXiv:2602.04120, XaaS treats explainability as a first-class system service rather than a model-specific feature. It introduces a distributed explanation cache with semantic similarity to allow edge devices to request, cache, and verify explanations under resource and latency constraints. This addresses the inefficiency of current coupled methods that generate explanations simultaneously with inferences, causing redundant computation and poor scalability across heterogeneous edge devices.