基于深度学习的复杂网络实时Sybil攻击检测算法

REAL-TIME SYBIL ATTACK DETECTION ALGORITHM FOR COMPLEX NETWORKS BASED ON DEEP LEARNING

针对复杂网络中Sybil攻击检测速度较慢的问题,提出一种基于深度学习的复杂网络实时Sybil攻击检测方案。从网络中采集数据,提取合适的特征;通过深度学习技术预测网络中的攻击行为。基于多层核极限学习机的深度学习技术包括无监督表示学习与监督特征分类两个阶段。通过低秩逼近法计算近似的经验核映射,代替原极限学习机随机生成的隐层。将经验核映射-自动编码的栈式自编码器作为表示学习,对极限学习机的时间效率与存储成本实现了显著的提高。基于实际社交数据的实验结果表明,该方案有效地降低了Sybil攻击的检测时间,并且保持了较好的检测效果。

Aiming at the problem of low speed of Sybil attacks detection in the complex networks,this paper proposed a real-time Sybil attack detection for complex networks based on deep learning. Firstly,the data was collected from the network,and the appropriate features were extracted. Then deep learning technique was used to predict the attack behaviors of networks. In this paper,deep learning technique based on multilayer kernel extreme learning machine consisted of unsupervised representation learning and supervised features classification. An approximate empirical kernel was computed through low rank approximation method in replace of the randomly generated hidden layer of original extreme learning machine. The stacked auto encoders of empirical kernel were used for representation learning,so that the time efficiency and the storage cost of extreme learning machine were both improved observably. Experimental results based on the real social data indicate that the proposed scheme reduces the detection time of Sybil attacks effectively,and it realizes good detection results too.