摘要 |
<P>PROBLEM TO BE SOLVED: To obtain a pattern learning model which has scalability and versatility. Ž<P>SOLUTION: Learning modules 10<SB>1</SB>to 10<SB>N</SB>respectively performs update learning to update a plurality of model parameters of a pattern learning model by using input data. A model parameter sharing part 20 causes the learning modules 10<SB>1</SB>to 10<SB>N</SB>to share the model parameters. As for each of the learning modules 10<SB>1</SB>to 10<SB>N</SB>, a sharing strength updating part 112 updates sharing strength β<SB>ij</SB>indicating the strength of influence from the learning module 10<SB>j</SB>to a learning module 10<SB>i</SB>so that a learning error E<SB>i</SB>of the learning module 10<SB>i</SB>in updating the plurality of model parameters by update learning can be minimized. This invention may be applied to the learning of, for example, time-sequential patterns. Ž<P>COPYRIGHT: (C)2010,JPO&INPIT Ž
|