Markov Graphs
本文用对数线性模型刻画具有一般依赖结构和马尔可夫依赖的随机图，证明马尔可夫图的充分统计量由各种三角形和星形计数给出，并讨论简单马尔可夫图的推断方法。
Abstract Log-linear statistical models are used to characterize random graphs with general dependence structure and with Markov dependence. Sufficient statistics for Markov graphs are shown to be given by counts of various triangles and stars. In particular, we show under which assumptions the triad counts are sufficient statistics. We discuss inference methodology for some simple Markov graphs. Key Words: Log-linear network modelMarkov fieldGraph inference