MILD

MILD framework: proactive multi-intent failure prediction and root-cause disambiguation for autonomous networks

Researchers introduce MILD, a framework that reframes network intent assurance from reactive drift detection to proactive failure prediction in self-driving (autonomous) networks. MILD uses a teacher-augmented Mixture-of-Experts architecture with a hybrid objective that jointly optimizes failure prediction and root-cause attribution across three operational macro-intents (telemetry, analytics, actuation). The system incorporates SHAP-based KPI-level diagnostics and multi-horizon urgency estimation, evaluated across statistical benchmarks, microservices environments, and an SDN-based edge-to-cloud testbed. The work is relevant to AI-driven infrastructure automation and closed-loop network management.