Hybridizing Evolutionary Computation and Deep Neural Networks: An Approach to Handwriting Recognition Using Committees and Transfer Learning

Table 2

Summary of errors (in %) of the best 20 GA and GE individuals after full training in MNIST. Source: Baldominos et al. [3].

	GA		GE
#	Mean	Best	Mean	Best
1	0.5130	0.41	0.4680	0.41
2	0.5380	0.49	0.4765	0.42
3	0.5485	0.49	0.5310	0.47
4	0.5875	0.50	0.4300	0.37
5	0.6115	0.46	0.5545	0.47
6	0.5010	0.45	0.5265	0.42
7	0.5760	0.50	0.4615	0.40
8	0.6085	0.53	0.5830	0.50
9	0.4795	0.40	0.5530	0.48
10	0.6090	0.57	0.5330	0.48
11	0.5600	0.49	0.4985	0.44
12	0.5850	0.54	0.4815	0.42
13	0.7095	0.60	0.5090	0.42
14	0.6045	0.54	0.5680	0.50
15	0.5780	0.47	0.5095	0.48
16	0.6265	0.51	0.5245	0.45
17	0.6010	0.51	0.5525	0.48
18	0.6815	0.62	0.6015	0.52
19	0.4805	0.40	0.5100	0.46
20	0.5930	0.52	0.5525	0.48