Byrne W
Dept. of Electr. Eng., Maryland Univ., College Park, MD.
IEEE Trans Neural Netw. 1992;3(4):612-20. doi: 10.1109/72.143375.
Training a Boltzmann machine with hidden units is appropriately treated in information geometry using the information divergence and the technique of alternating minimization. The resulting algorithm is shown to be closely related to gradient descent Boltzmann machine learning rules, and the close relationship of both to the EM algorithm is described. An iterative proportional fitting procedure for training machines without hidden units is described and incorporated into the alternating minimization algorithm.
使用信息散度和交替最小化技术，在信息几何中对具有隐藏单元的玻尔兹曼机进行训练是恰当的。结果表明，所得算法与梯度下降玻尔兹曼机学习规则密切相关，并且描述了两者与期望最大化（EM）算法的紧密关系。描述了一种用于训练无隐藏单元机器的迭代比例拟合过程，并将其纳入交替最小化算法中。
