Feature Engineering for Deep Reinforcement Learning Based Routing

摘要: Recent advances in Deep Reinforcement Learning (DRL) techniques are providing a dramatic improvement decision-making and automated control problems. As result, we witnessing growing number of research works that proposing ways applying DRL to network-related problems such as routing. However, proposals failed achieve good results, often under-performing traditional routing techniques. We argue successfully DRL-based networking requires finding representations the network parameters: feature engineering. agents need represent both state (e.g., link utilization) action space changes policy). In this paper, show existing approaches use straightforward lead poor performance. propose novel representation outperforms ones is flexible enough be applied many use-cases. test our two different scenarios: (i) optical transport networks (ii) QoS-aware IP networks. Our results agent achieves significantly better performance compared state/action representations.