发明名称 SYSTEM AND METHOD FOR JUDGING NEURAL NETWORK LEARNING LEVEL
摘要 PROBLEM TO BE SOLVED: To make easily recognizable and visualizable the learning states of a mutual connection type neural network (SOM) as levels by providing a learning controller of the SOM, an output log storage which stores the output positions and a learning level judging device which judges the learning levels of the SOM. SOLUTION: A learning controller 2 controls the learning of a SOM 1 self- organizingly. A learning level judging device 4 receives the output number and the block type number of an output log of an output log storage 3 as the inputs. The output positions and the block type number, i.e., the output results of past 10 times are stored in the storage 3. The device 4 refers to the block type number to which the current input belongs to count the number of blocks belonging to the block type number and judges the learning levels of the SOM 1 based on the counting results of blocks belonging to each block type number which are obtained for all of 10 inputs.
申请公布号 JP2000132533(A) 申请公布日期 2000.05.12
申请号 JP19980307197 申请日期 1998.10.28
申请人 NEC CORP 发明人 HATAKEYAMA AKEMI
分类号 G06F15/18;G06N3/08;G09B9/00;(IPC1-7):G06F15/18 主分类号 G06F15/18
代理机构 代理人
主权项
地址