Carlos Natalino

Explanation-based runtime verification intercepts erroneous ML decisions in optical networks, preserving automation while enhancing reliability.

Models can achieve robust performance across diverse optical network configurations with a novel joint learning approach that enhances adaptability and efficiency.

A medium-sized LLM outperforms larger models in delivering high-quality, efficient explanations for optical network automation tasks.

Domain-specific tools can deliver 90% accuracy in optical network management while slashing token usage by three times compared to their generic counterparts.

Human-understandable explanations for AI decisions can be achieved by augmenting traditional XAI methods with LLMs and mutual feature interactions, leading to significantly improved interpretability.

Achieving over 99% accuracy in anomaly detection and classification without retraining could revolutionize security protocols in optical networks.

Achieving over 83% improvement in cross-system accuracy for optical threat detection could redefine the robustness of ML models in diverse environments.