Smart Communication Networks with Machine Learning

在向第五代（5G）及以后发展的过程中，通信网络的设计和运营方式发生了根本性的变化。这种转变主要是由于不断增长的服务需求、新的用例和多样化的应用程序，这使得传统的“一刀切”方法变得无效和低效。软件定义网络（SDN）和网络功能虚拟化（NFV）是两种新兴技术，可用于实现智能、敏捷和经济高效的网络解决方案。由于现代通信网络的复杂操作，机器智能可能为处理数据流量动态和最大化网络性能提供新的途径。该DG项目将研究几个基本问题，利用最先进的机器智能工具，为智能和敏捷通信网络开发创新工程解决方案。首先，我们将学习如何确定虚拟网络的网络拓扑。拓扑的建立需要联合路由和业务功能的布局，是一个多目标优化问题，复杂度较高。我们将开发基于流量负载预测的动态网络拓扑解决方案。其次，我们将研究如何有效地分配多维（计算、传输和存储）资源，以便在共享公共物理基板的多个虚拟网络之间实现所需的服务隔离；在保证服务质量的同时，开发资源管理框架，实现资源的最大复用。最后，我们将研究可靠的端到端数据传输的网络协议定制。所需要的协议取决于与每个业务请求关联的网络和业务功能。因此，应该根据需要定制和创建协议。我们将开发一个协议定制和自动化解决方案。HQP培训是DG应用的重要组成部分。5名博士生和3名博士后将参与拟开展的研究。该研究计划将为学员提供通信网络和机器学习方面的宝贵知识和研究经验，并使他们成为加拿大高科技产业劳动力的专家补充。研究成果将对科学、技术和社会产生深远的影响，为未来通信网络开发具有成本效益的创新智能算法和协议提供深入的理解和新的见解，以支持各种新的用例和信息服务。毫无疑问，开发基于机器智能的网络技术的竞赛将由世界各地的研究团体进行。在这个问题上发明成功的新技术将使加拿大电信业在国际市场上具有重要的竞争优势。

In evolving towards the fifth generation (5G) and beyond, the communication networks have been experiencing a fundamental change in the way they are designed and operated. The shift is mainly due to the ever-increasing service demands, new use cases, and diverse applications, which render the traditional one-size-fits-all approach ineffective and inefficient. Software defined networking (SDN) and network function virtualization (NFV) are two emerging technologies that can be leveraged for a smart, agile, and cost-effective networking solution. Due to the complicate operation of modern communication networks, machine intelligence potentially provides a new avenue to deal with data traffic dynamics and to maximize network performance. This DG program will investigate several fundamental issues to develop innovative engineering solutions for smart and agile communication networks, taking advantage of state-of-the-art machine intelligence tools. First, we will study how to determine the network topology for a virtual network. The topology establishment requires joint routing and service function placement, which is a multi-objective optimization problem and entails high complexity. We will develop a dynamic network topology solution based on traffic load prediction. Second, we will investigate how to efficiently allocate multi-dimensional (computing, transmission, and storage) resources for desired service isolation among multiple virtual networks sharing the common physical substrate; and develop a resource management framework for maximal resource multiplexing while ensuring service quality. Lastly, we will study networking protocol customization for reliable end-to-end data transport. The required protocol depends on network and service functions associated with each service request. As a result, the protocol should be customized and created on demand. We will develop a protocol customization and automation solution. HQP training is an essential component of this DG application. Five PhD students and three postdoctoral fellows will participate in the proposed research. The research program will provide the trainees with valuable knowledge and research experience in communication networks and machine learning, and enable them to emerge as expert additions to the Canadian high technology industry labor force. The research outcomes are expected to have a profound scientific, technological, and social impact, in providing in-depth understanding and new insights for developing innovative cost-effective intelligent algorithms and protocols for future communication networks, to support various new use cases and information services. The race towards developing networking technologies based on machine intelligence will undoubtedly be undertaken by research groups worldwide. Invention of successful new techniques on the subject will foster a vital competitive edge to Canadian telecommunications industry in the international market place.