CCRI: Medium: Collaborative Research: 3DML: A Platform for Data, Design and Deployed Validation of Machine Learning for Wireless Networks and Mobile Applications

无线网络及其新兴用户应用（如自动驾驶汽车、虚拟现实和电子医疗）的复杂性不断增加，刺激了对开发机器学习（ML）支持的智能网络管理和优化的巨大需求。然而，释放这种创新仍然存在两个主要障碍：(i)基于机器学习的方法需要许多大型标记数据集，由于隐私和成本的挑战，这些数据集很难在无线环境中获得；（ii）将复杂的ML模型部署到资源受限的无线设备中的挑战。该项目的总体目标是设计、开发和传播一个名为3DML的社区平台，以促进下一代无线网络和移动应用中基于ml的创新的发展。3DML将是第一个从底层设计的平台，以满足探索基于ml的无线应用创新的迫切需求，具有三个集成的关键组件。首先，本项目将开发能够在不同规模的网络中运行并捕获不同网络运行状态的3DML-Data，并能够收集前所未有的多样化的标记数据集。其次，该项目将设计3DML-Client，它由自动化工具和压缩库组成，以(i)自动生成高效的ML模型和部署策略，以在给定的不同设备和应用程序之间实现任务性能和资源消耗的最佳权衡，以及（ii）为快速开发提供高效ML模块和功能的综合池。第三，该项目将开发3dml基础设施，它可以利用3dml客户端从3dml数据收集的数据，生成部署到无线基础设施中的高效ML算法，并包括一种方法，供研究人员使用ML算法定制大规模MIMO信道估计、检测、解码、波束形成和频谱共享的关键模块。该项目解决了无线研究社区开发基于机器学习的智能网络管理平台的迫切需求。该项目的成功将提供数据和工具，使基于机器学习的无线应用方法的自动化和自定义探索和部署成为可能。该教育项目包括研讨会、在线课程和实习，不仅包括来自不同学院的本科生和研究生，还包括来自行业的从业者。总体而言，3DML将为下一代智能无线网络的创新开发开辟一系列新的可能性，包括增强型移动宽带、大规模物联网和超低延迟应用，以支持众多新兴应用。所有开发的数据集、工具和库将在https://3dml.rice.eduThis上发布，奖励反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

The ever-increasing complexity of wireless networks and their emerging novel user applications (such as autonomous cars, virtual reality, and e-health) have spurred a significant demand to develop machine learning (ML) empowered, intelligent network management and optimization. However, there are still two main barriers to unleashing such innovations: (i) ML-based approaches require many large labeled datasets, which are difficult to acquire in the wireless context due to both privacy and cost challenges; and (ii) the challenge of deploying complex ML models into resource-constrained wireless devices. This project’s overarching goal is to design, develop, and disseminate a community platform called 3DML, for facilitating the development of ML-based innovations for next-generation wireless networks and mobile applications. 3DML will be the first platform, designed from the ground up, to meet the urgent need of exploring ML-based innovations for wireless applications, featuring three integrated key components. First, this project will develop 3DML-Data which has the ability to operate in networks with different scales and capture diverse network operating states and enable the collection of unprecedentedly diverse labeled datasets. Second, this project will design 3DML-Client, which consists of automated tools and compression libraries to (i) automatically generate efficient ML models and deployment strategies for achieving optimal trade-offs between task performance and resource consumption given diverse devices and applications, and (ii) provide a comprehensive pool of efficient ML modules and functions for fast development. Third, this project will develop 3DML-Infrastructure, which can make use of 3DML-Client with data collected from 3DML-Data, to generate efficient ML algorithms deployed into wireless infrastructure, and include a methodology for researchers to use ML algorithms to customize key modules for massive MIMO channel estimation, detection, decoding, beamforming, and spectrum sharing. This project addresses a pressing need of the wireless research community to develop a platform for ML-empowered intelligent network management. The success of this project will provide data and tools to enable automated and self-customized exploration and deployment of ML-based approaches for wireless applications. The educational program with workshops, online courses, and internships will involve not only undergraduate and graduate students from various institutes, but also practitioners from industry. Overall, 3DML will open up a host of new possibilities for developing innovations towards next generation intelligent wireless networks, including enhanced mobile broadband, massive Internet-of-things and ultra-low-latency applications in order to support numerous emerging applications. All of the developed datasets, tools, and libraries will be released at https://3dml.rice.eduThis award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.