Operations for learning with graphical models

摘要: This paper is a multidisciplinary review of empirical, statistical learning from graphical model perspective. Well-known examples models include Bayesian networks, directed graphs representing Markov chain, and undirected networks field. These are extended to data analysis empirical using the notation plates. Graphical operations for simplifying manipulating problem provided including decomposition, differentiation, manipulation probability exponential family. Two standard algorithm schemas reviewed in framework: Gibbs sampling expectation maximization algorithm. Using these schemas, some popular algorithms can be synthesized their specification. includes versions linear regression, techniques feed-forward Gaussian discrete frorn data. The concludes by sketching implications summarizing how fall within framework presented. The main original contributions here decomposition demonstration that provide understanding developing complex algorithms.