Deductive Reasoning for Joint Distribution Probability in Simple Topic Model
Deductive Reasoning for Joint Distribution Probability in Simple Topic Model
复制标题
DOI：
10.1109/iiai-aai.2016.31
复制
发表时间：
2016-07
期刊：
影响因子：
--
通讯作者：
Y. Shirota;T. Hashimoto;B. Chakraborty
中科院分区：
文献类型：
--
作者：
Y. Shirota;T. Hashimoto;B. Chakraborty
贝叶斯推理广泛应用于数据工程等各种应用领域。当我们推导后验时，我们必须结合许多定理或规则，例如贝叶斯定理。即使我们使用概率图模型，后验表达式的推导也是相当困难的。因此，我们为此提出了一种基于演绎推理的方法。文中给出了一个简单主题模型的具体推导图。演绎推理图阐明了哪些定理以及它们在演绎中的使用方式。另外，对后验推导中常用的三种条件独立模式规则进行了直观的解释。
Bayesian inference is widely used in various application field such as data engineering. When we derive the posterior, we have to combine many theorems or rules such as the Bayes' theorem. The derivation of the posterior expression is quite difficult, even if we use the probabilistic graphical model. So we propose a deductive reasoning based approach for that. The concrete deductive diagram for a simple topic model is presented in the paper. The deductive reasoning diagram clarifies which theorems and how they are used in the deduction. In addition, the three conditional independence pattern rules which are used frequently in the posterior derivation are explained visually.