水下激光探测系统隐蔽性分析

根据水下激光探测系统的光源特性、探测原理和应用背景,分析了可能对激光水下探测系统形成威胁的探测手段及其探测能力.采用Monte Carlo方法对激光水下探测系统的隐蔽性问题进行了计算和仿真研究,结果表明:降低探测光单脉冲能量、选择有一定环境照度的天气条件、加大水下激光探测系统载具的使用深度等手段可以极大地提高水下激光探测系统的隐蔽性与安全性.通过分析得到了不同的探测手段对激光水下探测系统的威胁程度,同时从理论上提出了针对不同探测手段的激光水下探测系统的安全应用深度.     

Base communication model of IP covert timing channels

IP covert timing channel (IPCTC) is an unconventional communication channel which attaches time information to the packets of an overt channel as messages carriers, e.g., using different inter-packet delays to transmit messages in a packet-switched network. Although the IPCTCs have many different communication methods, based on the concept of time, we categorized the base communication model of the IPCTCs into three types and then utilized the signal processing theory to build their mathematical models. As a result, the basic characteristics of the IPCTCs’ base model were formally derived. Hence, the characteristics of any IPCTC can be derived from the base models that consist of the IPCTC. Furthermore, a set of approaches was devised to implement the base model of the IPCTCs in a TCP/IP network. Experimental results show the correctness of the proposed base model of the IPCTCs in this paper.     

一种提高水印算法隐蔽性的区域测定规则

提出了一种基于层次模型的规则。规则中采用检测机制在文本宿主文件区域上得到检测值,并依据该检测值和极限值来确定嵌入水印信息的区域。依据该规则,提高了已有文本数字水印算法的隐蔽性。实验分析表明与该规则结合使用的算法抗攻击性与单独使用算法相比较有明显提高。     

基于显著图的高隐蔽性模型指纹算法

在核心任务场景下训练深度神经网络(DNN)需要越来越多的算力资源,这刺激了基于云端预测API接口的模型的窃取与盗用,同时也违反了模型所有者的知识产权.为了追踪公开的非法模型副本,深度神经网络的模型指纹技术为希望保持模型完整性的模型所有者提供了一种强大的版权验证方案.然而,现有的模型指纹方案主要基于输出层面的内在痕迹(例如:特定输入样本下的错误预测行为),这导致在模型指纹验证阶段缺乏隐蔽性.本文基于模型预测时的显著图(saliency map)痕迹,提出了一种全新的任意下游任务通用的模型指纹方案.本文的方案提出了受约束的显著图操控目标,构建标签不变和自然的指纹样本,显著提高了模型指纹的隐蔽性.根据对3种典型任务场景下全面的评估结果,本文提出的方法被证明能够显著地增强现有方案的指纹版权验证的效果,同时保持高度的模型指纹隐蔽性.