| 摘要 |
PROBLEM TO BE SOLVED: To construct an approximate value outputting device (CMAC) which outputs a value approximating a true value decided by a combination of inputted data with a small amount of data quantity and in a short time. SOLUTION: This CMAC has a cerebellum model part provided with a plurality of layers of layer models having a granule cell value in each combination of a plurality of sections obtained by dividing a range that can be taken by various data and a calculating part which specifies the granule cell value of a combination of corresponding sections from a plurality of kinds of inputted data, adds the granule cell values of each layer and calculates an approximate value, and the division position of the cerebellum model part is made suitable by evolutionary algorithm within the range of the number of designated divisions. This learning device for a granule cell value has a specifying part which specifies a granule cell value before learning corresponding to the combination of plurality of kinds of inputted data, an adding part which adds specified granule cell values, a deviation calculating part which calculates the deviation between the added value and the true value, an updating part which updates the granule cell value before learning on the basis of the deviation and the number of the layers of the layer model and a controlling part which starts the specification processing of the specifying part before the updating part completes update processing.
|
