Federated Learning for Energy-Efficient Task Computing in Wireless Networks
Federated Learning for Energy-Efficient Task Computing in Wireless Networks
复制标题
DOI：
10.1109/icc40277.2020.9148625
复制
发表时间：
2020-06
期刊：
影响因子：
--
通讯作者：
Sihua Wang;Mingzhe Chen;W. Saad;Changchuan Yin
中科院分区：
文献类型：
--
作者：
Sihua Wang;Mingzhe Chen;W. Saad;Changchuan Yin
本文研究了在具有移动的边缘计算（MEC）能力的蜂窝网络中最小化任务计算和传输能耗的问题。在所考虑的网络中，每个用户需要在每个时隙处理一个计算任务。任务的一部分可以被发送到基站（BS），基站可以使用其强大的计算能力来处理从其用户卸载的任务。由于每个用户的计算任务的数据大小随时间变化，BS必须动态地调整资源分配方案以满足用户的需求。该问题被视为一个优化问题，其目标是通过调整用户关联方案以及任务和功率分配方案来最小化任务计算和传输的能量消耗。针对这一问题，提出了一种基于支持向量机的联邦学习方法来主动确定用户关联。给定用户关联，BS可以收集与其关联用户的计算任务相关的信息，使用该信息，每个用户的发射功率和任务分配将被优化，并且每个用户的能量消耗也被最小化。所提出的基于SVM的FL方法使BS和用户能够合作建立全局SVM模型，该模型可以确定所有用户的关联，而无需任何用户的历史关联和计算任务卸载的传输。利用上海交通大学OMNILab的真实的城市蜂窝业务数据进行仿真，结果表明，与传统的集中式SVM方法相比，该算法可以降低用户的能耗高达20.1%.
In this paper, the problem of minimizing energy consumption for task computation and transmission in a cellular network with mobile edge computing (MEC) capabilities is studied. In the considered network, each user needs to process a computational task at each time slot. A part of the task can be transmitted to a base station (BS) that can use its powerful computational ability to process the tasks offloaded from its users. Since the data size of each user’s computational task varies over time, the BSs must dynamically adjust the resource allocation scheme to meet the users’ needs. This problem is posed as an optimization problem whose goal is to minimize the energy consumption for task computing and transmission via adjusting user association scheme as well as their task and power allocation scheme. To solve this problem, a support vector machine (SVM)-based federated learning (FL) is proposed to determine the user association proactively. Given the user association, the BS can collect the information related to the computational tasks of its associated users using which, the transmit power and task allocation of each user will be optimized and the energy consumption of each user is also minimized. The proposed SVM-based FL method enables the BS and users to cooperatively build a global SVM model that can determine all users’ association without any transmission of users’ historical association and computational task offloading. Simulations using real data on city cellular traffic from the OMNILab at Shanghai Jiao Tong University show that the proposed algorithm can reduce the users’ energy consumption by up to 20.1% compared to the conventional centralized SVM method.