CANs promise to apply cognition to overcome shortcomings of self-organizing networks, such as limited flexibility and adaptability to changing environments. in CAN, machine-learning-based network automation functions, called CFs, learn context-specific policies for automating network operations. For this, CFs need a common abstract description of the network states to which they respond. This article presents a design and implementation of an EMA engine that could be tasked with learning the required abstract states in a consistent way across multiple CFs.

中文翻译：

---

使用深度学习对网络和环境状态进行建模和抽象


CAN 有望应用认知来克服自组织网络的缺点，例如灵活性和对不断变化的环境的适应性有限。在 CAN 中，基于机器学习的网络自动化功能（称为 CF）可以学习特定于上下文的策略以实现网络操作自动化。为此，CF 需要对其响应的网络状态有一个通用的抽象描述。本文介绍了 EMA 引擎的设计和实现，该引擎的任务是跨多个 CF 以一致的方式学习所需的抽象状态。