Cooperation-Controlled Learning for Explicit Class Structure in Self-Organizing Maps

Table 1

Information, input information, quantization, and topographic errors obtained by the three methods with random initialization for the automobile data. The parameter was set to 100 to stabilize learning. In the table, “information” and “input inf” represent mutual information and input information, respectively.

Methods	Measures		Information	Input inf.	QE	TE
SOM	Average				0.068	0.040
	Std dev.				0.001	0.018
UCL	Average	10	4.354	0.000	0.048	0.125
	Std dev.		0.024	0.000	0.001	0.069
	Average	20	4.151	0.000	0.061	0.031
	Std dev.		0.036	0.000	0.001	0.022
	Average	30	4.099	0.000	0.064	0.030
	Std dev.		0.043	0.000	0.001	0.018
	Average	40	4.065	0.000	0.065	0.035
	Std dev.		0.044	0.000	0.001	0.014
	Average	50	4.044	0.000	0.066	0.035
	Std dev.		0.047	0.000	0.001	0.018
WCL	Average	10	4.241	0.554	0.030	0.088
	Std dev.		0.024	0.038	0.001	0.042
	Average	20	4.066	0.547	0.036	0.042
	Std dev.		0.035	0.021	0.001	0.019
	Average	30	3.861	0.560	0.041	0.028
	Std dev.		0.064	0.030	0.002	0.012
	Average	40	3.767	0.556	0.043	0.028
	Std dev.		0.046	0.03	0.002	0.014
	Average	50	3.749	0.556	0.043	0.025
	Std dev.		0.059	0.031	0.002	0.012