Graph Embedding Matrix Sharing With Differential Privacy
Graph Embedding Matrix Sharing With Differential Privacy
复制标题
具有差异隐私的图嵌入矩阵共享
DOI：
10.1109/access.2019.2927365
复制
发表时间：
2019-07
期刊：
影响因子：
3.9
通讯作者：
Ni Weiwei
中科院分区：
文献类型：
--
作者：
Zhang Sen;Ni Weiwei
图嵌入将图映射为低维向量，即嵌入矩阵，同时保持图的结构，解决了图分析的计算量和空间成本高的问题。矩阵分解（MF）是实现图嵌入的有效手段，因为它保持了图结构的实用性。嵌入矩阵中隐含的个性化图结构特征可以识别个体，这可能会泄露原始图中的个人敏感信息。目前，在不影响效用的前提下保护个人隐私是实现嵌入矩阵共享的关键。差分隐私是在保护隐私的同时发布敏感信息的黄金标准。然而，现有的差分私有MF方法由于具有较高的全局灵敏度或迭代噪声误差积累，无法直接应用到基于MF的图嵌入中，可能导致基于MF的图嵌入的实用性较差。为了解决这一不足，本研究提出了一种基于mf的图嵌入矩阵共享的差分私有摄动梯度下降方法PPGD。具体地说，设计了MF目标函数的Lipschitz条件和梯度裁剪策略来限定全局灵敏度。在此过程中，提出了一种独立于原始数据集的可扩展全局灵敏度解决方案。此外，在梯度下降中设计了一种复合噪声添加方法，在提高实用性的同时保证了私密性。理论分析表明，PPGD在实现$(\varepsilon, \delta)$ -差分隐私的同时，能够生成效用最大化的处理嵌入矩阵。实验验证了PPGD的有效性和效率。
Graph embedding maps a graph into low-dimensional vectors, i.e., embedding matrix, while preserving the graph structure, solving the high computation and space cost for graph analysis. Matrix factorization (MF) is an effective means to achieve graph embedding since maintaining the utility of the graph structure. The personalized graph structure features implied in the embedding matrix can identify the individual, which potentially breaches individual sensitive information in the original graph. Currently, protecting individual privacy without compromising the utility is the key to sharing the embedding matrix. Differential privacy is a gold standard for publishing sensitive information while protecting privacy. The existing methods on differentially private MF, however, cannot be directly incorporated onto MF-based graph embedding as they undergo either high global sensitivity or iterative noise error accumulation, potentially rendering poor utility of MF-based graph embedding. To address the deficiency, this study proposes PPGD, a differentially private perturbed gradient descent method for MF-based graph embedding matrix sharing. Specifically, a Lipschitz condition on the objective function of the MF and a gradient clipping strategy are devised for bounding global sensitivity. Along the way, a scalable solution to global sensitivity that is independent on the original dataset is proposed. Further, a composite noise added means in the gradient descent is designed to guarantee privacy while enhancing the utility. The theoretical analysis shows that PPGD can generate processed embedding matrix with the utility maximization while achieving $(\varepsilon, \delta)$ -differential privacy. The experimental evaluations confirm the effectiveness and efficiency of PPGD.
登录
查看更多内容
DOI：
--
发表时间：
2014-06
期刊：
ArXiv
影响因子：
--
作者：
Yann Dauphin;Razvan Pascanu;Çaglar Gülçehre;Kyunghyun Cho;S. Ganguli;Yoshua Bengio
通讯作者：
Yann Dauphin;Razvan Pascanu;Çaglar Gülçehre;Kyunghyun Cho;S. Ganguli;Yoshua Bengio
DOI：
10.5555/1953048.2021036
发表时间：
2009-11
期刊：
Journal of machine learning research : JMLR
影响因子：
--
作者：
Kamalika Chaudhuri;C. Monteleoni;A. Sarwate
通讯作者：
Kamalika Chaudhuri;C. Monteleoni;A. Sarwate
Jo-DPMF：通过联合优化进行差分私有矩阵分解学习
影响因子：
8.1
作者：
Zhang, Feng;Lee, Victor E.;Choo, Kim-Kwang Raymond
通讯作者：
Choo, Kim-Kwang Raymond
DOI：
10.1145/2792838.2800173
发表时间：
2015-09
期刊：
Proceedings of the 9th ACM Conference on Recommender Systems
影响因子：
--
作者：
Arnaud Berlioz;Arik Friedman;M. Kâafar;R. Boreli;S. Berkovsky
通讯作者：
Arnaud Berlioz;Arik Friedman;M. Kâafar;R. Boreli;S. Berkovsky
DOI：
10.1145/3219819.3220076
发表时间：
2018-07
期刊：
Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining
影响因子：
--
作者：
Jaewoo Lee;Daniel Kifer
通讯作者：
Jaewoo Lee;Daniel Kifer