论文部分内容阅读
本文研究在无标度先验下,图模型的结构学习问题.提出新的正则化模型,其惩罚项为Log型和Lq型惩罚函数的复合,该模型包含无标度先验.本文使用重赋权迭代算法求解该模型.实验表明,所提出的新模型有效、实用,其在参数估计和模型选择方面均有良好效果.
In this paper, we study the structural learning problem of graph model under scale-free priori and propose a new regularization model whose penalty term is compound of Log-type and Lq-type penalty functions, which contains scale-less priori. Empirically iterative algorithm to solve this model.Experiments show that the proposed new model is effective and practical and it has good effect on parameter estimation and model selection.