Classification

We evaluated LDNN extensively on many benchmark data including: general binary and multi-class datasets from UCI repository, synthetic datasets such as two-moon and spiral datasets and also popular computer vision datasets such as MNIST, CIFAR10, CIFAR 100 and SVHN.

Binary datasets

Dataset	Classifier	Av. test err. (%)
Adults	LDNN	15.25
	MLP	15.75
	RF	14.14
	SVM	15.57
Wis. Breast Cancer	LDNN	0.80
	MLP	2.28
	RF	1.79
	SVM	1.59
PIMA Diabetes	LDNN	17.92
	MLP	22.11
	RF	20.81
	SVM	21.57
Australian	LDNN	12.93
	MLP	15.65
	RF	12.95
	SVM	16.59
Ionosophere	LDNN	3.40
	MLP	12.10
	RF	5.38
	SVM	4.27
German Credit	LDNN	22.58
	MLP	26.96
	RF	24.28
	SVM	25.83
Forest Cov. Type	LDNN	8.87
	MLP	12.22
	RF	3.90
	SVM	6.91
IJCNN 2001	LDNN	1.28
	MLP	2.34
	RF	2.00
	SVM	1.41
COD-RNA	LDNN	3.36
	MLP	3.68
	RF	3.37
	SVM	3.67
Webspam	LDNN	1.21
	MLP	2.44
	RF	1.17
	SVM	0.78

Multi-class datasets

Dataset	Classifier	Av. test err (%)
Isolet	LDNN	4.17
	RF	5.61
	SVM	3.21
Landsat	LDNN	7.98
	RF	9.15
	SVM	8.15
Letter	LDNN	2.32
	RF	3.89
	SVM	2.35
Optdigit	LDNN	2.29
	RF	2.89
	SVM	1.56
Pendigit	LDNN	1.80
	RF	3.64
	SVM	1.86
MNIST	LDNN	1.23
	RF	3.00
	SVM	1.40

Scientific Computing and Imaging Institute