摘要
为增强网络系统在对抗性攻击环境下的安全防御能力,采用模型蒸馏方法构建健壮性增强机制,分析温度系数与损失权重等参数对防御效果的影响,并将其与多种主流防御策略进行对比。在典型网络安全数据集上开展实证研究,结果表明该方法在保障标准性能的同时,显著增强了对多类扰动样本的适应能力,具备良好的稳定性与应用推广价值。
In order to enhance the security defense capability of network systems in an adversarial attack environment,a model distillation method is used to construct a robustness enhancement mechanism,analyze the influence of parameters such as temperature coefficient and loss weight on the defense effect,and compare it with a variety of mainstream defense strategies.By carrying out empirical studies on typical network security datasets,the results show that the method significantly enhances the adaptability to multiple types of perturbation samples while guaranteeing the standard performance,and has good stability and application promotion value.
作者
彭一楠
PENG Yinan(Shanghai Palm Information Technology Co.,Ltd.,Shanghai 200011)
出处
《智能物联技术》
2025年第4期113-117,共5页
Technology of Io T& AI
关键词
模型蒸馏
对抗性攻击
网络系统
防御机制
model distillation
adversarial attack
network system
defense mechanism
