Collaborative Research: IMR: MM-1A: Functional Data Analysis-aided Learning Methods for Robust Wireless Measurements

随着大规模、异构、动态和复杂无线网络的日益增长，如何在5G及以后的网络中实现准确和鲁棒的测量成为一个具有挑战性和重要的问题。大多数现有的数据驱动解决方案都是黑盒方法，可能不够健壮和自适应，并且仅适用于低维和离散数据。实际上，无线数据属于函数数据类，可以用曲线或函数来表示。功能数据分析（FDA）可以更好地处理高维无线数据集。认识到上述问题的重要性，本项目旨在弥合基于fda的学习与无线测量之间的差距。建议的研究分为以下四个相互交织的重点。(i)稀疏无线测量的功能数据回归：开发基于深度学习的方法来解决功能数据的基本回归问题。（ii）基于fda的动态无线测量迁移学习：研究动态环境下有效无线测量测试数据与训练数据分布移位下功能数据回归与分类的迁移学习。（iii）基于分位数fda的鲁棒无线测量和控制学习：开发一种基于深度学习的方法，以解决基于分位数回归方法的根本瓶颈。（iv）集成和验证的无线测量应用。如果成功，本研究将极大地促进无线测量和相关领域功能数据的实践和理解。教育和外联部分包括：(i)利用学习理论和FDA加强课程，并联合开发一门基于FDA的无线测量学习研究生课程。（ii）让本科生参与实践项目。现有的外展项目将为本科生提供研究机会和研讨会，重点是吸引代表性不足的学生。提高公众认识的外联活动，包括期刊出版物、会议发言、研讨会、IEEE杰出讲座、期刊特刊、主要会议的讲习班和特别会议。该项目生成的代码将在公共存储库GitHub （https://github.com/）上发布。项目网站将在奥本大学维护，网址为：https://www.eng.auburn.edu/~szm0001/proj_lMR23.html。该项目网站将经常定期更新，以发布该项目的成果，包括项目描述、项目团队、主要成果（如出版物、代码和数据集），以及对美国国家科学基金会对该项目的支持表示感谢。在为期三年的项目期内，该网站将由PI管理/更新。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

With the increasing growth of large-scale, heterogeneous, dynamic, and complex wireless networks, how to achieve accurate and robust measurements in 5G networks and beyond becomes a challenging and important problem. Most existing data-driven solutions are black-box approaches, which may not be robust and adaptive, and work only for low-dimensional and discrete data. In fact, wireless data belong to the class of functional data, which can be represented by curves or functions. High-dimensional wireless datasets can be better handled by functional data analysis (FDA). Recognizing the significance of the aforementioned problems, this project aims to bridge the gap between FDA-based learning and wireless measurement. The proposed research falls into the following four interwoven thrusts. (i) Functional Data Regression for Sparse Wireless Measurements: to develop a deep learning based approach to address fundamental regression problems of functional data. (ii) FDA-based Transfer Learning for DynamicWireless Measurements: to study transfer learning for functional data regression and classification under the distribution shift between test data and training data for effective wireless measurements in dynamic environments. (iii) Quantile FDA-based Learning for Robust Wireless Measurements and Control: to develop a deep learning-based approach to address the fundamental bottleneck of quantile regression-based methods. (iv) Wireless Measurement Applications for Integration and Validation. If successful, this research will greatly advance the practice and understanding of functional data for wireless measurement and related fields. The educational and outreach components include: (i) Curriculum enhancement with learning theory and FDA, and joint developing a graduate course on FDA-based learning for wireless measurements. (ii) Engaging undergrads with hands-on projects. The existing outreach programs will be leveraged to offer research opportunities and seminars to undergrads, with emphasis on engaging underrepresented students. (iii) Outreach activities to increase public awareness, include journal publications, conference presentations, seminars, IEEE distinguished lectures, journal special issues, and workshops and special sessions at major conferences.The code produced from this project will be disseminated at the public repository GitHub (https://github.com/). A project website will be maintained at Auburn University with URL: https://www.eng.auburn.edu/~szm0001/proj_lMR23.html. This project website will be frequently and regularly updated for dissemination of the outcomes from this project, including a description of the project, project team, major outcomes such as publications, codes and datasets, as well as an acknowledgement of NSF support to this project. This website will be managed/updated by the PI for the three-year project period.This 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.