To perform relabeling so that a meaning of an intermediate node changed by learning by EM algorithm and meanings of the respective state items of the intermediate node approach the intention of an expert.
In a Bayesian network configuration, the intermediate node at the same distance from an output node or an input node, having the same front node and having the same number of state items is classified into the same layer and layers are selected in order from the one farthest from the output node or the one closest to the input node. Continuously, a node label and a state label of the intermediate node and its state item are determined so as to suit an intermediate node case most from among the node labels and the state labels showing the meanings of the intermediate nodes included in the selected layers, their state items, a CPT of the intermediate node and the preliminarily given intermediate node cases and the intermediate node, the node label and the state label of its state item are updated according to the determined node label and state label.
JP2005107748A | 2005-04-21 |
Hidetoshi Tachibana
Yasukazu Sato
Hiroshi Yoshimoto
Yasushi Kawasaki
Jun Okazawa
Next Patent: INFORMATION PROCESSOR AND METHOD FOR ASSIGNING MEMORY ADDRESS SPACE