Table 2.

Testing accuracy on MNIST data.

		Number of samples
Neurons	Method	0	2 K	5 K	10 K	20 K	50 K	100 K	200 K
100	CSNN	74.52	82.97	84.96	85.17	85.38	86.17	86.93	86.19
	CRBA	60.92	82.01	83.83	85.83	87.31	88.49	88.96	89.29
	CSNN+CRBA	72.92	83.10	84.23	85.30	86.27	86.96	87.20	87.49
400	CSNN	82.44	88.37	89.42	90.42	90.97	91.87	91.99	92.11
	CRBA	70.35	84.92	89.06	89.76	91.65	92.77	93.35	93.95
	CSNN+CRBA	79.87	87.02	88.81	89.90	90.93	91.86	92.45	92.91
1,600	CSNN	83.19	86.23	88.89	91.36	92.75	93.38	93.50	93.84
	CRBA	74.22	87.39	89.90	91.47	93.02	94.41	95.04	95.44
	CSNN+CRBA	78.30	86.76	89.42	90.86	92.76	94.00	95.01	95.48

Bold values denote the highest accuracies achieved at the specified number of training samples for each configuration (100, 400 and 1600 neurons).