Two-Dimensional Extreme Learning Machine

Table 3

Mean accuracy (± standard deviation) of fifty-trial comparison of ELM and 2DELM. Testing accuracy is also compared based on pairwise -tests at 95% significance level.

Dataset	ELM		2DELM
	Training	Testing	Training	Testing
Yale_32 × 32	1.0000 ± 0.0000	0.7578 ± 0.0585	1.0000 ± 0.0000	0.7880 ± 0.0524^•
Yale_64 × 64	1.0000 ± 0.0000	0.8473 ± 0.0454	1.0000 ± 0.0000	0.8720 ± 0.0346^•
ORL_32 × 32	1.0000 ± 0.0000	0.9333 ± 0.0233	1.0000 ± 0.0000	0.9416 ± 0.0244^•
ORL_64 × 64	1.0000 ± 0.0000	0.9251 ± 0.0258	1.0000 ± 0.0000	0.9378 ± 0.0275^•
UMist	1.0000 ± 0.0000	0.9649 ± 0.0139	1.0000 ± 0.0000	0.9749 ± 0.0128^•
GTface	1.0000 ± 0.0000	0.9752 ± 0.0150^∘	1.0000 ± 0.0000	0.9746 ± 0.0151
PIE	0.9766 ± 0.0016	0.9314 ± 0.0047	0.9888 ± 0.0009	0.9586 ± 0.0028^•
Letter	0.9543 ± 0.0011	0.9366 ± 0.0020^•	0.9489 ± 0.0009	0.9311 ± 0.0018
Shuttle	0.9995 ± 0.0000	0.9977 ± 0.0001^•	0.9989 ± 0.0001	0.9970 ± 0.0001
USPS	0.9910 ± 0.0006	0.9369 ± 0.0027	0.9918 ± 0.0007	0.9395 ± 0.0033^•
MNIST	0.9378 ± 0.0010	0.9357 ± 0.0016	0.9407 ± 0.0010	0.9403 ± 0.0016^•

• or ∘ indicates whether the advantage is significant or not.