Identity Access Management dataset

我们发布了使用商业公司的 IAM 图生成的 280 个合成 IAM 图。具体来说，我们改变节点的数量，但保持图形密度不变，即在 0.259 ± 0.198 (avg ± std) 的范围内。为了生成合成图，我们首先在以下区间 [10, 150] 和 [50, 300] 上分别从均匀分布中采样用户和数据存储的数量，这些区间分别涵盖了这些参数在真实图中的变化。在固定节点计数后，我们从随机选择的真实世界图中替换实际节点进行采样。然后我们将高斯 N(0, 0.01) 噪声添加到节点嵌入并重新归一化它们。为了使图密度与底层基线的密度相匹配，我们从多项分布中采样边缘，其中每个分量与用户和数据存储嵌入之间的余弦距离成正比。此外，我们强制执行动态边始终是所有权限边的子集的不变量。以这种方式生成的合成图是底层真实世界图的“上采样”版本。

We have released 280 synthetic IAM graphs generated from the IAM graphs provided by commercial entities. Specifically, we vary the number of nodes while maintaining a constant graph density within the range of 0.259 ± 0.198 (average ± standard deviation). To generate the synthetic graphs, we first sample the counts of users and data stores from uniform distributions over the intervals [10, 150] and [50, 300], respectively; these intervals cover the variations of these parameters in real-world graphs. Once the node count is fixed, we perform sampling by substituting actual nodes from randomly selected real-world graphs. Next, we add Gaussian noise N(0, 0.01) to the node embeddings and renormalize them. To match the graph density to that of the underlying baseline, we sample edges from a multinomial distribution, where each component is proportional to the cosine distance between the user and data store embeddings. Additionally, we enforce the invariant that dynamic edges always form a subset of all permission edges. The synthetic graphs generated in this manner are "upsampled" versions of the underlying real-world graphs.

搜集汇总

背景与挑战

背景概述

该数据集包含280个基于商业公司IAM图生成的合成IAM图，通过调整节点数量并保持图密度在0.259 ± 0.198范围内来模拟真实场景。生成过程涉及采样用户和数据存储数量、添加噪声处理节点嵌入，并确保动态边为权限边的子集，从而创建真实图的“上采样”版本，发布于2022年，相关研究论文可在arXiv获取。