The security problems of wireless sensor networks (WSN) have attracted people’s wide attention. In this paper, after we have summarized the existing security problems and solutions in WSN, we find that the insider at...The security problems of wireless sensor networks (WSN) have attracted people’s wide attention. In this paper, after we have summarized the existing security problems and solutions in WSN, we find that the insider attack to WSN is hard to solve. Insider attack is different from outsider attack, because it can’t be solved by the traditional encryption and message authentication. Therefore, a reliable secure routing protocol should be proposed in order to defense the insider attack. In this paper, we focus on insider selective forwarding attack. The existing detection mechanisms, such as watchdog, multipath retreat, neighbor-based monitoring and so on, have both advantages and disadvantages. According to their characteristics, we proposed a secure routing protocol based on monitor node and trust mechanism. The reputation value is made up with packet forwarding rate and node’s residual energy. So this detection and routing mechanism is universal because it can take account of both the safety and lifetime of network. Finally, we use OPNET simulation to verify the performance of our algorithm.展开更多
New precisely cooperative attacks, such as the coordi- nated cross plane session termination （CXPST） attack, need thou- sands upon thousands machines to attack diverse selected links simultaneously with the given ra...New precisely cooperative attacks, such as the coordi- nated cross plane session termination （CXPST） attack, need thou- sands upon thousands machines to attack diverse selected links simultaneously with the given rate. However, almost all command and control（C＆C） mechanisms only provide publishing one com- mand to the whole once, so-called one-to-all C＆C model, and are not productive to support CXPST-alike attacks. In this paper, we present one-to-any C＆C model on coordination among the unco- operative controlled nodes. As an instance of one-to-any C＆C model, directional command publishing （DCP） mechanism lever- aging on Kademlia is provided with a range-mapping key creating algorithm for commands to compute the publishing range and a statistically stochastic node querying scheme to obtain the com- mands immediately. With theoretical analysis and simulation, it is indicated that one-to-any C＆C model fits for precisely coordi- nated operation on uncooperative controlled nodes with least complexity, better accuracy and efficiency. Furthermore, DCP mechanism can support one-to-all command publishing at the same time. As an example of future C＆C model, studying on one-to-any C＆C model may help to promote the development of more efficient countermeasures.展开更多
图垂直联邦学习是一种结合图数据和垂直联邦学习的分布式机器学习方法,广泛应用于金融服务、医疗健康和社交网络等领域。该方法在保护隐私的同时,利用数据多样性显著提升模型性能。然而,研究表明图垂直联邦学习容易受到对抗攻击的威胁...图垂直联邦学习是一种结合图数据和垂直联邦学习的分布式机器学习方法,广泛应用于金融服务、医疗健康和社交网络等领域。该方法在保护隐私的同时,利用数据多样性显著提升模型性能。然而,研究表明图垂直联邦学习容易受到对抗攻击的威胁。现有的针对图神经网络的对抗攻击方法,如梯度最大化攻击、简化梯度攻击等方法在图垂直联邦框架中实施时仍然面临攻击成功率低、隐蔽性差、在防御情况下无法实施等问题。为应对这些挑战,提出了一种面向图垂直联邦的对抗攻击方法(Node and Feature Adversarial Attack,NFAttack),该方法分别设计了节点攻击策略与特征攻击策略,从不同维度实施高效攻击。首先,节点攻击策略基于度中心性指标评估节点的重要性,通过连接一定数量的虚假节点以形成虚假边,从而干扰高中心性节点。其次,特征攻击策略在节点特征中注入由随机噪声与梯度噪声构成的混合噪声,进而扰乱分类结果。最后,在6个数据集和3种图神经网络模型上进行实验,结果表明NFAttack的平均攻击成功率达到80%,比其他算法提高了约30%。此外,即使在多种联邦学习防御机制下,NFAttack仍展现出较强的攻击效果。展开更多
文摘The security problems of wireless sensor networks (WSN) have attracted people’s wide attention. In this paper, after we have summarized the existing security problems and solutions in WSN, we find that the insider attack to WSN is hard to solve. Insider attack is different from outsider attack, because it can’t be solved by the traditional encryption and message authentication. Therefore, a reliable secure routing protocol should be proposed in order to defense the insider attack. In this paper, we focus on insider selective forwarding attack. The existing detection mechanisms, such as watchdog, multipath retreat, neighbor-based monitoring and so on, have both advantages and disadvantages. According to their characteristics, we proposed a secure routing protocol based on monitor node and trust mechanism. The reputation value is made up with packet forwarding rate and node’s residual energy. So this detection and routing mechanism is universal because it can take account of both the safety and lifetime of network. Finally, we use OPNET simulation to verify the performance of our algorithm.
基金Supported by the National Natural Science Foundation of China(61402526,61502528)
文摘New precisely cooperative attacks, such as the coordi- nated cross plane session termination （CXPST） attack, need thou- sands upon thousands machines to attack diverse selected links simultaneously with the given rate. However, almost all command and control（C＆C） mechanisms only provide publishing one com- mand to the whole once, so-called one-to-all C＆C model, and are not productive to support CXPST-alike attacks. In this paper, we present one-to-any C＆C model on coordination among the unco- operative controlled nodes. As an instance of one-to-any C＆C model, directional command publishing （DCP） mechanism lever- aging on Kademlia is provided with a range-mapping key creating algorithm for commands to compute the publishing range and a statistically stochastic node querying scheme to obtain the com- mands immediately. With theoretical analysis and simulation, it is indicated that one-to-any C＆C model fits for precisely coordi- nated operation on uncooperative controlled nodes with least complexity, better accuracy and efficiency. Furthermore, DCP mechanism can support one-to-all command publishing at the same time. As an example of future C＆C model, studying on one-to-any C＆C model may help to promote the development of more efficient countermeasures.
文摘图垂直联邦学习是一种结合图数据和垂直联邦学习的分布式机器学习方法,广泛应用于金融服务、医疗健康和社交网络等领域。该方法在保护隐私的同时,利用数据多样性显著提升模型性能。然而,研究表明图垂直联邦学习容易受到对抗攻击的威胁。现有的针对图神经网络的对抗攻击方法,如梯度最大化攻击、简化梯度攻击等方法在图垂直联邦框架中实施时仍然面临攻击成功率低、隐蔽性差、在防御情况下无法实施等问题。为应对这些挑战,提出了一种面向图垂直联邦的对抗攻击方法(Node and Feature Adversarial Attack,NFAttack),该方法分别设计了节点攻击策略与特征攻击策略,从不同维度实施高效攻击。首先,节点攻击策略基于度中心性指标评估节点的重要性,通过连接一定数量的虚假节点以形成虚假边,从而干扰高中心性节点。其次,特征攻击策略在节点特征中注入由随机噪声与梯度噪声构成的混合噪声,进而扰乱分类结果。最后,在6个数据集和3种图神经网络模型上进行实验,结果表明NFAttack的平均攻击成功率达到80%,比其他算法提高了约30%。此外,即使在多种联邦学习防御机制下,NFAttack仍展现出较强的攻击效果。
