摘要： 针对观测变量中含隐变量的非高斯线性无环因果模型的估计问题，提出一种新的算法。通过在超完备基独立成分分析算法中引入满足Oracle性质的惩罚因子，使混合矩阵的估计值具有稀疏连接权值，由此推导出模型估计算法。实验结果表明，该算法能够改进因果模型估计的精确程度，提高算法效率。

关键词: 线性因果模型, 超完备基, 独立成分分析, 稀疏连接

Abstract: In order to estimate the non-Gaussian linear acyclic causal models with latent variables, a new algorithm is proposed. The algorithm is derived by integrating penalty functions which satisfy the Oracle property into overcomplete Independent Component Analysis(ICA), making the entries of mixing matrix sparse. Experimental results show that the algorithm improves accuracy and efficiency of the casual model estimation.

Key words: linear causal model, overcomplete basis, Independent Component Analysis(ICA), sparse connection

中图分类号: 

• TP312