A Multikernel-Like Learning Algorithm Based on Data Probability Distribution

Table 5

Error rates (%) of the multiclass problem experiments on five datasets: USPS, MNIST, ISOLET, Yale-B, and CMU-PIE, where the value in bracket is the number of labeled sample in each class.

Dataset	Algorithm	Unlabeled (labeled samples)			Test (labeled samples)
USPS	MKDPD	19.8241 ( )	12.7056 ( )	10.3949 ( )	18.1100 ( )	10.9359 ( )	8.5658 ( )
	MR	23.5377 ()	14.5279 ()	11.2359 ()	21.2428 ()	13.3242 ()	10.2856 ()
MNIST	MKDPD	31.8342 ( )	20.5939 ( )	16.0872 ( )	38.5650 ( )	27.2950 ( )	23.0300 ( )
	MR	35.1206 ()	21.2741 ()	17.9026 ()	44.6050 ()	31.3800 ()	26.7300 ()
ISOLET	MKDPD	18.5743 ( )	12.7060 ( )	10.3704 ( )	29.9469 ( )	24.4297 ( )	21.7241 ( )
	MR	20.6300 ()	15.1305 ()	13.0057 ()	30.1790 ()	25.4111 ()	22.4801 ()
Yale-B	MKDPD	36.4440 ( )	26.6991 ( )	18.4750 ( )	35.3516 ( )	26.8203 ( )	17.2070 ( )
	MR	45.5172 ()	36.9676 ()	31.3500 ()	41.9531 ()	36.1406 ()	25.5078 ()
CMU-PIE	MKDPD	23.9773 ( )	11.5591 ( )	5.6034 ( )	20.1968 ( )	9.3056 ( )	4.1551 ( )
	MR	25.2904 ()	12.2715 ()	6.3793 ()	21.0532 ()	11.0069 ()	5.7292 ()