基于后验概率的Markov逻辑网参数学习方法

通过介绍统计关系学习方法Markov逻辑网的理论模型和参数学习方法, 提出一种基于后验概率的参数估计方法, 该方法采用正态先验分布, 用伪似然概率替代似然概率, 通过最大化伪后验概率来学习模型参数. 实验结果表明, 该方法能够有效地学出模型参数, 且所得模型推理能力优于现有的参数学习方法.

Pseudo-posterior Parameters Learning of Markov Logic Networks

The theory and parameters learning of MLNs are introduced,and a parameter learning method based on posterior is proposed.With normal distribution as the prior and pseudo likelihood instead of(likelihood),the pseudo-posterior is maximized to learn parameters.Experimental results show MLNs parameters can be effectively learned,and the inference with the learned model is better that those with current parameter learning methods.