| 摘要 |
<P>PROBLEM TO BE SOLVED: To provide a pattern learning model having 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 means 20 causes the learning modules 10<SB>1</SB>to 10<SB>N</SB>to share the model parameters. When new learning data are supplied to a pattern learning part 111 as input data, a module creation part 121 creates a new learning module corresponding to the new learning data. Then, after performing the update learning over all learning modules including the new learning module by all the learning modules, a module integration part 123 decides whether to integrate the learning modules based on similarities among the learning modules by a similarity evaluation part 122, and integrates the learning modules. This invention may be applied to the learning of, for example, time series patterns. Ž<P>COPYRIGHT: (C)2010,JPO&INPIT Ž
|
