6G联邦边缘学习新范式：基于任务导向的资源管理策略

doi:10.11959/j.issn.1000-436x.2022128

通信学报 ›› 2022, Vol. 43 ›› Issue (6): 16-27.doi: 10.11959/j.issn.1000-436x.2022128

• 专题：面向6G的智能至简网络关键技术 • 上一篇下一篇

6G联邦边缘学习新范式：基于任务导向的资源管理策略

王志勤¹, 江甲沫¹, 刘沛西², 曹晓雯³, 李阳¹, 韩凯峰¹, 杜滢¹, 朱光旭⁴

¹ 中国信息通信研究院移动通信创新中心，北京 100191
² 北京大学电子学院，北京 100871
³ 广东工业大学信息工程学院，广东广州 510006
⁴ 深圳市大数据研究院，广东深圳 518172

修回日期:2022-04-27 出版日期:2022-06-01 发布日期:2022-06-01
作者简介:王志勤（1970- ），女，北京人，博士，中国信息通信研究院教授级高级工程师，主要研究方向为无线移动通信技术和标准
江甲沫（1985- ），男，吉林长春人，博士，中国信息通信研究院高级工程师，主要研究方向为面向6G的无线人工智能、通信感知一体化技术
刘沛西（1991- ），男，湖北仙桃人，北京大学博士生，主要研究方向为面向6G的无线人工智能技术
曹晓雯（1995- ），女，广东河源人，广东工业大学博士生，主要研究方向为面向6G的无线人工智能、无线空中计算技术
李阳（1993- ），男，河南濮阳人，中国信息通信研究院助理工程师，主要研究方向为新一代移动通信与人工智能结合技术
韩凯峰（1993- ），男，北京人，博士，中国信息通信研究院高级工程师，主要研究方向为面向6G的无线人工智能、通信感知一体化技术
杜滢（1978- ），女，山东菏泽人，中国信息通信研究院教授级高级工程师，主要研究方向为无线移动通信技术和标准
朱光旭（1989- ），男，广东广州人，博士，深圳市大数据研究院研究员，主要研究方向为面向6G的无线人工智能技术
基金资助:
国家重点研发计划基金资助项目(2020YFB1807100);国家自然科学基金资助项目(62001310);广东省基础与应用基础研究基金资助项目(2022A1515010109)

New design paradigm for federated edge learning towards 6G:task-oriented resource management strategies

Zhiqin WANG¹, Jiamo JIANG¹, Peixi LIU², Xiaowen CAO³, Yang LI¹, Kaifeng HAN¹, Ying DU¹, Guangxu ZHU⁴

¹ Mobile Communication Innovation Center, China Academy of Information and Communications Technology, Beijing 100191, China
² School of Electronics, Peking University, Beijing 100871, China
³ School of Information Engineering, Guangdong University of Technology, Guangzhou 510006, China
⁴ Shenzhen Research Institute of Big Data, Shenzhen 518172, China

Revised:2022-04-27 Online:2022-06-01 Published:2022-06-01
Supported by:
The National Key Research and Development Program of China(2020YFB1807100);The National Natural Science Foundation of China(62001310);Guangdong Basic and Applied Basic Research Foundation(2022A1515010109)

摘要/Abstract

摘要：

目的：为充分利用分布在网络边缘的丰富数据使之服务于人工智能模型训练，以联邦边缘学习为代表的边缘智能技术应运而生，本文综述了面向 6G 的联邦边缘学习技术，能够充分利用分布在网络边缘的丰富数据使之服务于人工智能模型训练，并以最优化学习性能（如优化模型训练时间、学习收敛性等）为目标设计无线资源管理策略。

方法：利用联邦学习网络架构，本文对资源分配和用户调度方案进行分析：1.对于资源分配问题分析在以最小化总训练时间为目标时的通信轮数和每轮时延之间的折中关系，为了满足每轮训练的时间约束，应该给计算能力低的设备分配更多的频率带宽，通过减小通信时间补偿计算时间，反之亦然。因此，设备间的带宽分配应该同时考虑信道条件和计算资源，这与传统只考虑信道条件的带宽分配工作截然不同。为此，需要建模总训练时间最小化问题，优化量化级数和带宽分配，设计交替优化算法解决该问题。2.对于用户调度问题通过将数据重要性和仍需通信轮次相联系、信道质量和单轮通信延时相联系，利用理论模型将两者统一，建模通信时间最小化优化问题。通过求解该优化问题发现，最优的调度策略在前期会更多地注重数据重要性，而在后期更注重信道质量。此外，还将所提的单设备调度算法也扩展至多设备调度场景当中。

结果：1.对于资源分配问题，当带宽分配最优时，通过仿真得到的总训练时间与量化级数的关系，在每个量化级数上运行相同的训练过程至少5次，且存在使总训练时间最小化的最优量化级数。总训练时间 $T = N_{ϵ} \cdot T_{d}$ ，且 $N_{ϵ}$ 是量化水平 q 的递减函数，而T_d是 q 的递增函数。此外，通过理论优化中得到的最优量化级数与仿真结果一致，同时验证了所提算法的有效性。根据损失函数最优值间隔与训练时间的关系所示，通过求解训练时间最小化问题得到了最优量化级数和最优带宽分配策略。2.对于用户调度问题，仿真中将所提用户调度（TLM）方案与其他3个常见的调度方案比较，并分别显示了在通信时间为6000s和14000s时的平均精度，其中平均精度通过测量预测值与真实值的联合交叉（IoU, intersection of union）得到。其中CA方案在信道衰减最大的1号车产生了最差的精度，而 IA 方案在数据不太重要的4号车展现了最低的精度。ICA方案的目的是在CA和IA之间找到一个平衡点，但由于其启发式的性质，性能低于TLM方案。

结论：1.最优量化级数和最优带宽分配下的训练损失在更短的时间内达到了预定的阈值，并且能取得最高的测试准确率。其次，非最优量化级数和最优带宽分配下的训练性能会优于最优量化级数和平均带宽分配的性能，这同时验证了资源分配的必要性。2.TLM方案在训练早期取得了稍好的性能，充分训练后明显优于所有其他方案。这是由于所提TLM方案中固有的前瞻性质比现有的CA、IA和ICA方案中的近视性质更有优势。

关键词: 6G智能网络, 任务导向的联邦边缘学习, 资源管理策略, 用户调度策略

Abstract:

Objectives: To make full use of the abundant data distributed at the edge of the network to serve the training of artificial intelligence models, edge intelligence technology represented by federated edge learning emerges as the times require. The rich data at the edge enables it to serve artificial intelligence model training, and design wireless resource management strategies with the goal of optimizing learning performance (such as optimizing model training time, learning convergence, etc.).

Methods: Using the federated learning network architecture, this paper analyzes the resource allocation and user scheduling schemes: 1. For the resource allocation problem, the trade-off relationship between the number of communication rounds and the delay per round when the goal of minimizing the total training time is analyzed, To meet the time constraints of each round of training, more frequency bandwidth should be allocated to devices with low computational power, compensating for a computational time by reducing communication time, and vice versa. Therefore, bandwidth allocation between devices should consider both channel conditions and computing resources, which is completely different from the traditional bandwidth allocation work that only considers channel conditions. To this end, it is necessary to model the total training time minimization problem, optimize the quantization series and bandwidth allocation, and design an alternate optimization algorithm to solve this problem. 2. For the user scheduling problem, the communication time minimization optimization problem is modeled by linking the importance of data with the number of communication rounds, channel quality and single-round communication delay, and using a theoretical model to unify the two. By solving the optimization problem, it is found that the optimal scheduling strategy will pay more attention to the importance of data in the early stage, and pay more attention to the channel quality in the later stage. In addition, the proposed single-device scheduling algorithm is also extended to multi-device scheduling scenarios.

Results: 1. For the resource allocation problem, when the bandwidth allocation is optimal, the relationship between the total training time and the quantization level obtained by simulation, run the same training process at least 5 times on each quantization level, and there is a total training time. The total training time is $T = N_{ϵ} \cdot T_{d}$ , and $N_{ϵ}$ is a decreasing function of quantization level q,and T_d is an increasing function of q . In addition, the optimal quantization series obtained through theoretical optimization is consistent with the simulation results, and the effectiveness of the proposed algorithm is verified. According to the relationship between the optimal value interval of the loss function and the training time, the optimal quantization series and the optimal bandwidth allocation strategy are obtained by solving the training time minimization problem. 2. For the user scheduling problem, the proposed user scheduling (TLM) scheme is compared with three other common scheduling schemes in the simulation, and the average precision is shown when the communication time is 6 000 s and 14 000 s, where the average Accuracy is obtained by measuring the IoU, the intersection of union, of the predicted value and the true value. The CA scheme yields the worst accuracy on car 1 with the largest channel attenuation, while the IA scheme exhibits the lowest accuracy on car 4 where the data is less important. The ICA scheme aims to find a balance between CA and IA, but due to its heuristic nature, the performance is lower than that of the TLM scheme.

Conclusions: 1. The training loss under the optimal quantization level and optimal bandwidth allocation reaches the predetermined threshold in a shorter time and can achieve the highest test accuracy. Secondly, the training performance under the non-optimal quantization level and optimal bandwidth allocation will be better than the performance of the optimal quantization leveland average bandwidth allocation, which also verifies the necessity of resource allocation. 2. The TLM scheme achieves slightly better performance early in training and significantly outperforms all other schemes after full training. This is due to the inherent prospective nature in the proposed TLM protocol which is advantageous over the myopic nature in the existing CA, IA and ICA protocols.

Key words: intelligent network towards 6G, task-oriented federated edge learning, resource management strategy, user scheduling strategy

中图分类号:

TN929.5

王志勤, 江甲沫, 刘沛西, 曹晓雯, 李阳, 韩凯峰, 杜滢, 朱光旭. 6G联邦边缘学习新范式：基于任务导向的资源管理策略[J]. 通信学报, 2022, 43(6): 16-27.

Zhiqin WANG, Jiamo JIANG, Peixi LIU, Xiaowen CAO, Yang LI, Kaifeng HAN, Ying DU, Guangxu ZHU. New design paradigm for federated edge learning towards 6G:task-oriented resource management strategies[J]. Journal on Communications, 2022, 43(6): 16-27.

图/表 10

图1

图2

图3

图4

图5

图6

图7

图8

图9

图10

参考文献 51

[1]	IMT-2030（6G）推进组. 6G总体愿景与潜在关键技术白皮书[R]. 2021.
	IMT-2030 (6G) Promotion Group. White paper on 6G overall vision and potential key technology[R]. 2021.
[2]	张平, 牛凯, 田辉 ,等. 6G 移动通信技术展望[J]. 通信学报, 2019,40(1): 141-148.
	ZHANG P , NIU K , TIAN H ,et al. Technology prospect of 6G mobile communications[J]. Journal on Communications, 2019,40(1): 141-148.
[3]	LETAIEF K B , CHEN W , SHI Y ,et al. The roadmap to 6G:AI empowered wireless networks[J]. IEEE Communication Magazine, 2019,57(8): 84-90.
[4]	WANG Z Q , DU Y , WEI K J ,et al. Vision,application scenarios,and key technology trends for 6G mobile communications[J]. Science China Information Sciences, 2022,65(5): 1-27.
[5]	ZHU G X , LIU D Z , DU Y Q ,et al. Toward an intelligent edge:wireless communication meets machine learning[J]. IEEE Communications Magazine, 2020,58(1): 19-25.
[6]	LI E , ZENG L , ZHOU Z ,et al. Edge AI:On-demand accelerating deep neural network inference via edge computing[J]. IEEE Transaction on Wireless Communication, 2020,19(1): 447-457.
[7]	ZHU G X , XU J , HUANG K B ,et al. Over-the-air computing for wireless data aggregation in massive IoT[J]. IEEE Wireless Communications, 2021,28(4): 57-65.
[8]	KONECNY J , MCMAHAN H B , YU F X ,et al. Federated learning:strategies for improving communication efficiency[J]. arXiv Preprint,arXiv:1610.05492, 2016.
[9]	YANG Q , LIU Y , CHEN T ,et al. Federated machine learning:concept and applications[J]. ACM Transaction on Intelligent Systems, 2019,10(2): 1-19.
[10]	CHEN M , YANG Z , SAAD W ,et al. A joint learning and communications framework for federated learning over wireless networks[J]. IEEE Transactions on Wireless Communications, 2020,20(1): 269-283.
[11]	DAYAN I , ROTH H R , ZHONG A X ,et al. Federated learning for predicting clinical outcomes in patients with COVID-19[J]. Nature Medicine, 2021,27(10): 1735-1743.
[12]	SAVAZZI S , NICOLI M , BENNIS M ,et al. Opportunities of federated learning in connected,cooperative,and automated industrial systems[J]. IEEE Communications Magazine, 2021,59(2): 16-21.
[13]	MILLS J , HU J , MIN G Y . Communication-efficient federated learning for wireless edge intelligence in IoT[J]. IEEE Internet of Things Journal, 2020,7(7): 5986-5994.
[14]	LAN Q , WEN D Z , ZHANG Z Z ,et al. What is semantic communication? A view on conveying meaning in the era of machine intelligence[J]. arXiv Preprint,arXiv:2110.00196, 2021.
[15]	TSE D , VISWANATH P . Fundamentals of wireless communication[M]. Cambridge: Cambridge University Press, 2005.
[16]	SHEN C , XU J , ZHENG S H ,et al. Resource rationing for wireless federated learning:concept,benefits,and challenges[J]. IEEE Communications Magazine, 2021,59(5): 82-87.
[17]	CHEN M Z , YANG Z H , SAAD W ,et al. A joint learning and communications framework for federated learning over wireless networks[J]. IEEE Transactions on Wireless Communications, 2021,20(1): 269-283.
[18]	WANG Y M , XU Y Q , SHI Q J ,et al. Quantized federated learning under transmission delay and outage constraints[J]. IEEE Journal on Selected Areas in Communications, 2022,40(1): 323-341.
[19]	ZHU G X , WANG Y , HUANG K B . Broadband analog aggregation for low-latency federated edge learning[J]. IEEE Transactions on Wireless Communications, 2020,19(1): 491-506.
[20]	LUO R Y , TIAN H , NI W L . Communication-aware path design for indoor robots exploiting federated deep reinforcement learning[C]// Proceedings of 2021 IEEE 32nd Annual International Symposium on Personal,Indoor and Mobile Radio Communications. Piscataway:IEEE Press, 2021: 1197-1202.
[21]	WAN S , LU J X , FAN P Y ,et al. Convergence analysis and system design for federated learning over wireless networks[J]. IEEE Journal on Selected Areas in Communications, 2021,39(12): 3622-3639.
[22]	CHEN M Z , POOR H V , SAAD W ,et al. Convergence time optimization for federated learning over wireless networks[J]. IEEE Transactions on Wireless Communications, 2021,20(4): 2457-2471.
[23]	REN J Y , SUN J S , TIAN H ,et al. Joint resource allocation for efficient federated learning in Internet of Things supported by edge computing[C]// Proceedings of 2021 IEEE International Conference on Communications Workshops. Piscataway:IEEE Press, 2021: 1-6.
[24]	MCMAHAN B , MOORE E , RAMAGE D ,et al. Communication-efficient learning of deep networks from decentralized data[C]// Artificial Intelligence and Statistics. New York:PMLR, 2017: 1273-1282.
[25]	YANG H H , LIU Z Z , QUEK T Q S ,et al. Scheduling policies for federated learning in wireless networks[J]. IEEE Transactions on Communications, 2020,68(1): 317-333.
[26]	TAIK A , MLIKA Z , CHERKAOUI S . Data-aware device scheduling for federated edge learning[J]. IEEE Transactions on Cognitive Communications and Networking, 2022,8(1): 408-421.
[27]	WANG S Q , TUOR T , SALONIDIS T ,et al. Adaptive federated learning in resource constrained edge computing systems[J]. IEEE Journal on Selected Areas in Communications, 2019,37(6): 1205-1221.
[28]	RIZK E , VLASKI S , SAYED A H . Optimal importance sampling for federated learning[C]// Proceedings of 2021 IEEE International Conference on Acoustics,Speech and Signal Processing. Piscataway:IEEE Press, 2021: 3095-3099.
[29]	AMIRI M M , GüNDüZ D , . Federated learning over wireless fading channels[J]. IEEE Transactions on Wireless Communications, 2020,19(5): 3546-3557.
[30]	KONECNY J , MCMAHAN H B , YU F X ,et al. Federated learning:strategies for improving communication and efficiency[J]. arXiv Preprint,arXiv:1610.05492, 2016.
[31]	AMIRI M M , GUNDUZ D , KULKARNI S R ,et al. Federated learning with quantized global model updates[J]. arXiv Preprint,arXiv:2006.10672, 2020.
[32]	NORI M K , YUN S , KIM I M . Fast federated learning by balancing communication trade-offs[J]. IEEE Transactions on Communications, 2021,69(8): 5168-5182.
[33]	唐伦, 汪智平, 蒲昊 ,等. 基于自适应梯度压缩的高效联邦学习通信机制研究[J]. 电子与信息学报， 2022:doi.org/10.11999/JEIT211262.
	TANG L , WANG Z P , PU H ,et al. Research on efficient federated learning communication mechanism based on adaptive gradient compression[J]. Journal of Electronics ＆ Information Technology， 2022:doi.org/10.11999/JEIT211262.
[34]	BOUZINIS P S , DIAMANTOULAKIS P D , KARAGIANNIDIS G K . Wireless quantized federated learning:a joint computation and communication design[J]. arXiv Preprint,arXiv:2203.05878, 2022.
[35]	ZHU G X , DU Y Q , GüNDüZ D , ,et al. One-bit over-the-air aggregation for communication-efficient federated edge learning:design and convergence analysis[J]. IEEE Transactions on Wireless Communications, 2021,20(3): 2120-2135.
[36]	ALISTARH D , GRUBIC D , LI J ,et al. QSGD:communication-efficient SGD via gradient quantization and encoding[C]// Advances in Neural Information Process.Systems. Massachusetts:MIT Press, 2017: 1707-1718.
[37]	LIU P X , JIANG J M , ZHU G X ,et al. Training time minimization for federated edge learning with optimized gradient quantization and bandwidth allocation[J]. arXiv Preprint,arXiv:2112.14387, 2021.
[38]	RAZAVIYAYN M . Successive convex approximation:analysis and applications[D]. TwenCities:University of Minnesota, 2014.
[39]	AMIRI M M , GüNDüZ D , KULKARNI S R ,et al. Convergence of update aware device scheduling for federated learning at the wireless edge[J]. IEEE Transactions on Wireless Communications, 2021,20(6): 3643-3658.
[40]	REN J K , HE Y H , WEN D Z ,et al. Scheduling for cellular federated edge learning with importance and channel awareness[J]. IEEE Transactions on Wireless Communications, 2020,19(11): 7690-7703.
[41]	YANG H H , ARAFA A , QUEK T Q S ,et al. Age-based scheduling policy for federated learning in mobile edge networks[C]// Proceedings of 2020 IEEE International Conference on Acoustics,Speech and Signal Processing. Piscataway:IEEE Press, 2020: 8743-8747.
[42]	NISHIO T , YONETANI R . Client selection for federated learning with heterogeneous resources in mobile edge[C]// Proceedings of 2019 IEEE International Conference on Communications. Piscataway:IEEE Press, 2019: 1-7.
[43]	贺文晨, 郭少勇, 邱雪松 ,等. 基于DRL的联邦学习节点选择方法[J]. 通信学报, 2021,42(6): 62-71.
	HE W , GUO S , QIU X ,et al. Node selection method in federated learning based on deep reinforcement learning[J]. Journal on Communications, 2021,42(6): 62-71.
[44]	ZHANG M J , ZHU G X , WANG S ,et al. Accelerating federated edge learning via optimized probabilistic device scheduling[C]// Proceedings of 2021 IEEE 22nd International Workshop on Signal Processing Advances in Wireless Communications. Piscataway:IEEE Press, 2021: 606-610.
[45]	DOSOVITSKIY A , ROS G , CODEVILLA F ,et al. CARLA:an open urban driving simulator[J]. arXiv Preprint,arXiv:1711.03938, 2017.
[46]	YAN Y , MAO Y X , LI B . SECOND:sparsely embedded convolutional detection[J]. Sensors, 2018,18(10): 3337.
[47]	ZHANG Z J , WANG S , HONG Y C ,et al. Distributed dynamic map fusion via federated learning for intelligent networked vehicles[C]// Proceedings of 2021 IEEE International Conference on Robotics and Automation. Piscataway:IEEE Press, 2021: 953-959.
[48]	ABARI O , RAHUL H , KATABI D . Over-the-air function computation in sensor networks[J]. arXiv Preprint,arXiv:1612.02307, 2016.
[49]	JEON S , WANG C , GASTPAR M . Computation over Gaussian networks with orthogonal components[J]. IEEE Transaction on Information Theory, 2014,60(12): 7841-7861.
[50]	YANG K , JIANG T , SHI Y M ,et al. Federated learning via over-the-air computation[J]. IEEE Transactions on Wireless Communications, 2020,19(3): 2022-2035.
[51]	IMT-2030（6G）推进组. 通信感知一体化技术研究报告[R]. 2021.
	IMT-2030 (6G) Promotion Group. Research report on integrated sensing and communications technology[R]. 2021.

6G联邦边缘学习新范式：基于任务导向的资源管理策略

New design paradigm for federated edge learning towards 6G:task-oriented resource management strategies

在线阅读

PDF下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 10

参考文献 51

相关文章 15

Metrics

推荐阅读 0

[1]	杨龙, 赵丽, 周雨晨, 贺冰涛, 陈健. 缓存辅助的协作NOMA携能传输[J]. 通信学报, 2023, 44(6): 77-89.
[2]	王再见, 谷慧敏. 基于联合优化的网络切片资源分配策略[J]. 通信学报, 2023, 44(5): 234-245.
[3]	苏新, 张桂福, 行鸿彦, Zenghui Wang. 基于平衡生成对抗网络的海洋气象传感网入侵检测研究[J]. 通信学报, 2023, 44(4): 124-136.
[4]	蒋丽, 谢胜利, 田辉. 面向数字孪生边缘网络的区块链分片及资源自适应优化机制[J]. 通信学报, 2023, 44(3): 12-23.
[5]	余雪勇, 邱礼翔, 宋家宁, 朱洪波. 无人机辅助边缘计算中安全通信与能效优化策略[J]. 通信学报, 2023, 44(3): 45-54.
[6]	江沸菠, 彭于波, 董莉. 面向6G的深度图像语义通信模型[J]. 通信学报, 2023, 44(3): 198-208.
[7]	龙隆, 刘子辰, 陆在旺, 张玉成, 李蕾. 移动边缘网络下服务缓存与资源分配联合优化策略[J]. 通信学报, 2023, 44(1): 64-74.
[8]	廖勇, 程港, 李玉杰. 基于深度展开的大规模MIMO系统CSI反馈算法[J]. 通信学报, 2022, 43(12): 77-88.
[9]	辜方林, 黄育侦, 赵莺, 赵海涛, 魏急波. 基于单载波频域均衡的直接序列扩频自适应抗干扰技术[J]. 通信学报, 2022, 43(11): 26-34.
[10]	陈赓, 齐书虎, 沈斐, 曾庆田. 面向B5G多业务场景基于D3QN的双时间尺度网络切片算法[J]. 通信学报, 2022, 43(11): 213-224.
[11]	丛玉良, 孙闻晞, 薛科, 钱志鸿, 陈绵书. 基于改进的混合遗传算法的车联网任务卸载策略研究[J]. 通信学报, 2022, 43(10): 77-85.
[12]	苏新, 孟蕾蕾, 周一青, CELIMUGE Wu. 基于深度强化学习的海洋移动边缘计算卸载方法[J]. 通信学报, 2022, 43(10): 133-145.
[13]	刘彩霞, 季新生, 邬江兴. 移动通信网的内生安全共性问题及破解之道[J]. 通信学报, 2022, 43(9): 70-79.
[14]	朱晓荣, 陈康. 基于细粒度切片的6G网络弹性切换算法研究[J]. 通信学报, 2022, 43(9): 148-156.
[15]	许文俊, 吴思雷, 王凤玉, 林兰, 李国军, 张治. 基于多智能体强化学习的大规模灾后用户分布式覆盖优化[J]. 通信学报, 2022, 43(8): 1-16.