Stateless ARE: Action Recommendation Engine without Network State Measurement

Abstract

Network Operation Centers (NOC) are responsible for a communication network's efficient operation, traffic engineering, failure management, and network assurance. Due to the size and complexity of today's networks, the traditionally manual operations in an NOC are becoming more difficult to perform optimally. In response to that, in our previous work we showed that Artificial Intelligence (AI) can be utilized for autonomous action recommendation in an NOC. While in that work the network's state was measured, in this work we study if actions can be recommended without measuring the network's state, saving both time and processing power, reducing complexity, and avoiding mistakes in measuring network state. To that end, we design an AI-based action recommender that recommends actions for an NOC without first measuring the network state. Results show that although such a stateless action recommender does not initially outperform its stateful equivalent, it does significantly outperform a static network, leading to the conclusion that with more optimization and/or by choosing better AI methods, a stateless action recommender could potentially reach the same performance of a stateful action recommender.