YeeKal
ml

10_graph_model

YeeKal
"#ml"

probabilistic graphical model

  • vertex
  • edge
  • isomorphism [,aɪsə'mɔrfɪzm]
  • directed graph
  • undirected graph
  • weight: weight/length/cost of edges
  • graphical model
    • bayesian network, 贝叶斯网络(directed acyclic graphical model(DAG),有向无环图`)
    • markov random field, 马尔可夫随机场, undirected

bayesian network

贝叶斯网络中边代表连接关系,点代表在n个与之有连接关系的父节点同时出现的条件下的条件概率。

D-separation:

  • head-to-head
    • c未知的条件下,a,b独立

head2head

  • tail-to-tail
    • c已知的条件下,a,b独立

tail2tail

  • head-to-tail

    • c已知的条件下,a,b独立
    • 在较长的链式网络中,若$x_i$固定,则$x_{i+1}$与$x_i$之前的节点都独立。即$x_{i+1}$的状态只与前一个节点有关。这种顺次演变的随机过程,就叫做马尔科夫链(Markov chain).

    head2tail

markov network

MRF:马可夫无向图/马可夫随机场,markov random field

reference

  1. 从贝叶斯方法谈到贝叶斯网络