基于深度确定性策略梯度的虚拟网络功能迁移优化算法

针对NFV/SDN架构下,服务功能链(SFC)的资源需求动态变化引起的虚拟网络功能(VNF)迁移优化问题,该文提出一种基于深度强化学习的VNF迁移优化算法。首先,在底层CPU、带宽资源和SFC端到端时延约束下,建立基于马尔可夫决策过程(MDP)的随机优化模型,该模型通过迁移VNF来联合优化网络能耗和SFC端到端时延。其次,由于状态空间和动作空间是连续值集合,提出一种基于深度确定性策略梯度(DDPG)的VNF智能迁移算法,从而得到近似最优的VNF迁移策略。仿真结果表明,该算法可以实现网络能耗和SFC端到端时延的折中,并提高物理网络的资源利用率。

     

基于改进深度强化学习的虚拟网络功能部署优化算法

针对网络功能虚拟化/软件定义网络 (NFV/SDN)架构下,网络服务请求动态到达引起的服务功能链(SFC)部署优化问题,该文提出一种基于改进深度强化学习的虚拟网络功能(VNF)部署优化算法。首先,建立了马尔科夫决策过程 (MDP)的随机优化模型,完成SFC的在线部署以及资源的动态分配,该模型联合优化SFC部署成本和时延成本,同时受限于SFC的时延以及物理资源约束。其次,在VNF部署和资源分配的过程中,存在状态和动作空间过大,以及状态转移概率未知等问题,该文提出了一种基于深度强化学习的VNF智能部署算法,从而得到近似最优的VNF部署策略和资源分配策略。最后,针对深度强化学习代理通过ε贪婪策略进行动作探索和利用,造成算法收敛速度慢等问题,提出了一种基于值函数差异的动作探索和利用方法,并进一步采用双重经验回放池,解决经验样本利用率低的问题。仿真结果表示,该算法能够加快神经网络收敛速度,并且可以同时优化SFC部署成本和SFC端到端时延。     

Machine learning-based clustering protocols for Internet of Things networks: An overview

The Internet of Things (IoT) continues to expand the current Internet, opening the door to a wide range of novel applications. The increasing volume of the IoT requires effective strategies to overcome its challenges. Machine Learning (ML) has led to a growing technology that enables computers to solve problems without the need for knowledge of their intricate details. Over the past years, various ML techniques have been used to efficiently manage IoT networks. Clustering is a technique that has proven its performance in the networking domain. Many works in the literature have studied ML-based clustering methods for IoT networks, including their main properties, characteristics, underlying technologies, and open issues. In this paper, we focus on topology-centered ML-based clustering protocols for IoT networks. Specifically, we investigate the potential benefits of adopting the clustering approach to address several IoT challenges. Moreover, we provide a comprehensive taxonomy of ML-based clustering algorithms for IoT networks. Finally, we statistically analyze the incorporation of ML techniques for clustering in various IoT systems and highlight the related open issues.     

基于上下文信息与注意力特征的欺骗语音检测

随着语音合成和语音转换技术的快速发展，欺骗语音检测方法仍存在欺骗检测准确率低、通用性差等问题。因此，提出一种基于上下文信息与注意力特征的端到端的欺骗检测方法。该方法基于深度残差收缩网络（DRSN），利用双分支上下文信息协调融合模块（DCCM）聚集丰富的上下文信息，融合基于协调时频注意力机制（CTFA）的特征以获得具有上下文信息的跨维度交互特征，从而最大化捕获伪影的潜力。与最佳基线系统相比，在ASVspoof 2019 LA数据集中，所提方法在EER和t-DCF性能指标上分别降低68%和65%；在ASVspoof 2021 LA数据集中，所提方法的EER和t-DCF分别为4.81和0.311 5，分别降低48%和10%。实验结果表明，所提方法能有效提高欺骗语音检测的准确率和泛化能力。     

Floop: An efficient video coding flow for unmanned aerial vehicles

Under limited transmission conditions, many factors affect the efficiency of video transmission. During the flight of an unmanned aerial vehicle (UAV), frequent network switching often occurs, and the channel transmission condition changes rapidly, resulting in low-video transmission efficiency. This paper presents an efficient video coding flow for UAVs working in the 5G nonstandalone network and proposes two bit controllers, including time and spatial bit controllers, in the flow. When the environment fluctuates significantly, the time bit controller adjusts the depth of the recursive codec to reduce the error propagation caused by excessive network inference. The spatial bit controller combines the spatial bit mask with the channel quality multiplier to adjust the bit allocation in space to allocate resources better and improve the efficiency of information carrying. In the spatial bit controller, a flexible mini graph is proposed to compute the channel quality multiplier. In this study, two bit controllers with end-to-end codec were combined, thereby constructing an efficient video coding flow. Many experiments have been performed in various environments. Concerning the multi-scale structural similarity index and peak signal-to-noise ratio, the performance of the coding flow is close to that of H.265 in the low bits per pixel area. With an increase in bits per pixel, the saturation bottleneck of the coding flow is at the same level as that of H.264.