ID	原文	译文
4003	分析和总结它们在解决无线通信问题时的原理、适用性、设计方法和优缺点；	Based on the overview, the principle, applicability, design me-thodology, pros and cons on applying AI technologies to solve wireless communication problems were analyzed andsummarized.
4004	最后围绕存在的局限性指出智能无线通信技术的未来发展趋势和研究方向，期望为无线通信领域的后续研究提供帮助和参考。	Regarding the existed limitations, the future development trends and research directions on intelligent wire-less communication technologies were pointed out, to hopefully provide useful help and reference for the future researchin this field.
4005	针对城市环境成为移动通信、交通调度、疾病防控等领域的典型场景，而用户的移动行为建模对这些关键领域有重要的应用与研究价值，梳理、总结城市环境下移动行为建模的研究进展与现状，为该领域的相关研究提供文献概述。	Urban environment has become a typical scenario for areas of mobile communication, transportation schedul-ing, disease controlling and so on, and modelling user's mobility behavior plays an important role in these key applica-tions. The research development in this area was combed and summarized, which provided a literature review for relatedworks.
4006	首先讨论了城市环境移动行为建模问题面临的主要挑战及对应的核心科学问题，即移动行为数据增强算法、城市结构感知的移动行为模式识别、多时空尺度的移动行为预测模型和移动数据隐私保护机制问题。	Firstly, the main challenges in urban mobility modelling were discussed as well as the corresponding key scientif-ic problems, which included mobility data augmentation, urban structure-aware mobility behavior discovering, mul-ti-scale mobility behavior prediction and mobility data privacy protection.
4007	进一步地，围绕这些核心科学问题梳理总结了该领域近年来的发展脉络与最新研究成果，为未来的研究工作奠定了基础。	Furthermore, according to these key scientificproblems, the recent developments and up-to-date scientific output in this area were summarized, which paved the wayfor future research.
4008	针对无线随机接入网络中通信信道状态信息（C-CSI）和干扰信道状态信息（I-CSI）均不确定的情况，提出了一种深度稳健资源分配架构。	A deep and robust resource allocation framework was proposed for the random access based wireless networks,where both the communication channel state information (C-CSI) and the interference channel state information (I-CSI)were uncertain.
4009	该资源分配架构将无线网络的资源优化目标看作一个学习问题，利用深度神经网络（DNN）以无监督的方式学习最优资源分配策略。	The proposed resource allocation framework considered the optimization objective of wireless networksas a learning problem and employs deep neural network (DNN) to approximate optimal resource allocation policythrough unsupervised manner.
4010	通过将不确定的 CSI 建模为椭圆形状的不确定性集合，提出了一个由 2 个 DNN 级联构成的网络结构，第一个是不确定的 CSI 处理单元，第二个是功率控制单元。	By modeling the uncertainties of CSI as ellipsoid sets, two concatenated DNN units wereproposed, where the first was uncertain CSI processing unit and the second was the power control unit.
4011	然后，提出了一种交替迭代训练算法用于联合训练 2 个级联的 DNN 单元。	Then, an alternat-ing iterative training algorithm was developed to jointly train the two concatenated DNN units.
4012	最后，仿真比较了稳健学习策略和非稳健学习策略下的网络性能，验证了所提算法的有效性。	Finally, the simulationsverify the effectiveness of the proposed robust leaning approach over the nonrobust one.