互联网时延稀疏估计中恶意节点的贪婪性识别

为了增强基于坐标的互联网时延稀疏估计算法遭受恶意攻击的生存能力,提出了一种分布式环境下的恶意节点识别算法。攻击者总是试图以尽可能小的代价实现显著的攻击效果。在分别利用l_1和l_2损失函数计算坐标以进行时延估计时,这种贪婪特征体现为估计误差的显著差异。分别以SMACOF和增量次梯度下降法代入不同损失函数计算临时坐标,利用给定阈值,清洗在不同临时坐标下的估计误差差异过大的参考节点;并二次代入l_1损失函数计算最终坐标。仿真实验证明,该方法能够在不影响时延估计精度的前提下,实现对恶意节点的有效识别。

Greedy Feature Identification against Malicious Nodes in Internet Latency Sparse Estimation

To enhance the survival of coordinate-based Internet latency sparse estimation methods while attacks are launched, a malicious nodes classification algorithm in full decentralized environment is presented. Attackers always try to get theShighest possible effects of theSattacksSatStheSlowestSpossibleScost. The greedy nature is shown the significantSdifferences between the estimation errors while the l₁ and l₂ loss functions is adopted specifically to get the coordinates for latency estimating. By introducing SMACOF and incremental sub-gradient descending algorithms to optimizing the l₁ and l₂ loss functions severally, a couple of temporary coordinates can be obtained. While this couple of temporary coordinates are both adopted, the malicious nodes can be rejected by a specific threshold for their excessive estimation error difference levels. Thus other benign reference nodes can be accepted to get a final coordinate by optimizing the l₁ loss function again. Simulations and experiments show that malicious can be identified effectively without losing latency estimation accuracy.