英语翻译 Another general approach to handle an intra

英语翻译

Another general approach to handle an intractable posterior distributionis to use sampling
methods to estimate the posteriorapproximately. For graphical models inimage labeling,
Markov Chain Monte Carlo (MCMC) sampling,including Gibbs sampling and Metropolis-
Hastings sampling,are widely used inpractice.The general MCMC methods sample a distri-
bution by constructing a Markov chainhaving the target distribution as its equilibrium distri-
bution.The state of the chain after alarge number of steps is used as a sample from the target
distribution.In particular,Gibbs sampling[21] assumes that we can sample each label variable
given its neighbors' configuration.Itrepeatedly scans all the variables,and each step in a cycle
consists of taking a sample from theposterior distribution of a variable given the values of all
others 图片为准 不用全部翻译 到文本这里
希望人工翻译!

人工翻译,请审阅
处理一个棘手的后验分布的另一种一般的方法是采用采样方法来近似估算后验部分.对于图像标签中的图形模型来说,马尔科夫链蒙特卡罗（MCMC）采样,包括Gibbs采样和Metropolis-Hastings采样,在实践中被广泛应用.一般MCMC方法采样一个分布是通过构建一个马尔可夫链,该马尔可夫链将目标分布作为其平衡分布.在大量步骤之后的链的状态被用作来自目标分布的一个样本.特别是,Gibbs采样[21]假设我们可以给出其邻居（相邻标签?）的配置的每个标签的变量.它重复地扫描所有的变量,并且在一个周期中的每个步骤都包括从给出所有其它值的变量的后验分布中取一个样本.