当前位置: X-MOL 学术Mobile Netw. Appl. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Differential Privacy Enabled Deep Neural Networks for Wireless Resource Management
Mobile Networks and Applications ( IF 2.3 ) Pub Date : 2022-08-19 , DOI: 10.1007/s11036-022-02013-6
Md Habibur Rahman , Md Munjure Mowla , Shahriar Shanto

Deep neural networks (DNN) are increasingly utilized for wireless resource allocations in beyond 5G/6G networks to solve the high computational time problem of iterative algorithms. The main issue of neural network-based wireless resource allocation schemes is that it is possible to regain sensitive details about the training data from model parameters. However, existing works do not consider the privacy leakage issues of the neural networks while allocating wireless resources. To resolve this problem, we develop a framework using two DNN architectures, e.g., multi-layer perceptron (MLP) network and convolutional neural network (CNN) based on the concept of differential privacy (DP) which is usually implemented for data privacy protection based on neural networks incorporating appropriately calibrated noise to reduce the sensitivity of the gradients. The results of the numerical simulation indicate that the DP-enabled CNN performs better achievable rate compared to DP-enabled MLP. Yet, the proposed framework solves the high computational time problem of the iterative algorithm, i.e., stochastic weighted minimum mean square error (SWMMSE). Evaluation illustrates that our proposed framework facilitates the design of privacy-enabled resource management in different sized wireless networks.



中文翻译:

用于无线资源管理的差分隐私启用深度神经网络

深度神经网络 (DNN) 越来越多地用于 5G/6G 网络之外的无线资源分配,以解决迭代算法的高计算时间问题。基于神经网络的无线资源分配方案的主要问题是可以从模型参数中重新获得有关训练数据的敏感细节。然而,现有的工作在分配无线资源时没有考虑神经网络的隐私泄露问题。为了解决这个问题,我们开发了一个使用两种 DNN 架构的框架,例如,基于差分隐私 (DP) 概念的多层感知器 (MLP) 网络和卷积神经网络 (CNN) 通常用于基于神经网络的数据隐私保护,该神经网络结合适当校准的噪声以降低梯度的敏感性。数值模拟的结果表明,与支持 DP 的 MLP 相比,支持 DP 的 CNN 具有更好的可实现率。然而,所提出的框架解决了迭代算法的高计算时间问题,即随机加权最小均方误差(SWMMSE)。评估表明,我们提出的框架有助于在不同规模的无线网络中设计启用隐私的资源管理。数值模拟的结果表明,与支持 DP 的 MLP 相比,支持 DP 的 CNN 具有更好的可实现率。然而,所提出的框架解决了迭代算法的高计算时间问题,即随机加权最小均方误差(SWMMSE)。评估表明,我们提出的框架有助于在不同规模的无线网络中设计启用隐私的资源管理。数值模拟的结果表明,与支持 DP 的 MLP 相比,支持 DP 的 CNN 具有更好的可实现率。然而,所提出的框架解决了迭代算法的高计算时间问题,即随机加权最小均方误差(SWMMSE)。评估表明,我们提出的框架有助于在不同规模的无线网络中设计启用隐私的资源管理。

更新日期:2022-08-20
down
wechat
bug