从众多用户收集的高维数据可用性越来越高,庞大的高维数据涉及用户个人隐私,如何在使用高维数据的同时保护用户的隐私极具挑战性。文中主要关注本地差分隐私下的高维数据发布问题。现有的解决方案首先构建概率图模型,生成输入数据的一...从众多用户收集的高维数据可用性越来越高,庞大的高维数据涉及用户个人隐私,如何在使用高维数据的同时保护用户的隐私极具挑战性。文中主要关注本地差分隐私下的高维数据发布问题。现有的解决方案首先构建概率图模型,生成输入数据的一组带噪声的低维边缘分布,然后使用它们近似输入数据集的联合分布以生成合成数据集。然而,现有方法在计算大量属性对的边缘分布构建概率图模型,以及计算概率图模型中规模较大的属性子集的联合分布时存在局限性。基于此,提出了一种本地差分隐私下的高维数据发布方法PrivHDP(High-dimensional Data Publication Under Local Differential Privacy)。首先,该方法使用随机采样响应代替传统的隐私预算分割策略扰动用户数据,提出自适应边缘分布计算方法计算成对属性的边缘分布构建Markov网。其次,使用新的方法代替互信息度量成对属性间的相关性,引入了基于高通滤波的阈值过滤技术缩减概率图构建过程的搜索空间,结合充分三角化操作和联合树算法获得一组属性子集。最后,基于联合分布分解和冗余消除,计算属性子集上的联合分布。在4个真实数据集上进行实验,结果表明,PrivHDP算法在k-way查询和SVM分类精度方面优于同类算法,验证了所提方法的可用性与高效性。展开更多
Over the past decade, open-source software use has grown. Today, many companies including Google, Microsoft, Meta, RedHat, MongoDB, and Apache are major participants of open-source contributions. With the increased us...Over the past decade, open-source software use has grown. Today, many companies including Google, Microsoft, Meta, RedHat, MongoDB, and Apache are major participants of open-source contributions. With the increased use of open-source software or integration of open-source software into custom-developed software, the quality of this software component increases in importance. This study examined a sample of open-source applications from GitHub. Static software analytics were conducted, and each application was classified for its risk level. In the analyzed applications, it was found that 90% of the applications were classified as low risk or moderate low risk indicating a high level of quality for open-source applications.展开更多
文摘从众多用户收集的高维数据可用性越来越高,庞大的高维数据涉及用户个人隐私,如何在使用高维数据的同时保护用户的隐私极具挑战性。文中主要关注本地差分隐私下的高维数据发布问题。现有的解决方案首先构建概率图模型,生成输入数据的一组带噪声的低维边缘分布,然后使用它们近似输入数据集的联合分布以生成合成数据集。然而,现有方法在计算大量属性对的边缘分布构建概率图模型,以及计算概率图模型中规模较大的属性子集的联合分布时存在局限性。基于此,提出了一种本地差分隐私下的高维数据发布方法PrivHDP(High-dimensional Data Publication Under Local Differential Privacy)。首先,该方法使用随机采样响应代替传统的隐私预算分割策略扰动用户数据,提出自适应边缘分布计算方法计算成对属性的边缘分布构建Markov网。其次,使用新的方法代替互信息度量成对属性间的相关性,引入了基于高通滤波的阈值过滤技术缩减概率图构建过程的搜索空间,结合充分三角化操作和联合树算法获得一组属性子集。最后,基于联合分布分解和冗余消除,计算属性子集上的联合分布。在4个真实数据集上进行实验,结果表明,PrivHDP算法在k-way查询和SVM分类精度方面优于同类算法,验证了所提方法的可用性与高效性。
文摘Over the past decade, open-source software use has grown. Today, many companies including Google, Microsoft, Meta, RedHat, MongoDB, and Apache are major participants of open-source contributions. With the increased use of open-source software or integration of open-source software into custom-developed software, the quality of this software component increases in importance. This study examined a sample of open-source applications from GitHub. Static software analytics were conducted, and each application was classified for its risk level. In the analyzed applications, it was found that 90% of the applications were classified as low risk or moderate low risk indicating a high level of quality for open-source applications.
