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

Table 3

Side-by-side comparison of the results for the EMNIST dataset along with the reported accuracy, including works using similar datasets from NIST Special Database 19.

Technique	Letters	Digits
Linear classifier [43]	55.78%	84.70%
OPIUM [43]	85.15%	95.90%
SVMs (one against all + sigmoid) [51]	–	98.75%
Multi-layer perceptron [53]	–	98.39%
Hidden Markov model [56]	90.00%	98.00%
Record-to-record travel [54]	93.78%	96.53%
PSO + fuzzy ARTMAP NNs [52]	–	96.49%
Multi-layer perceptron [55]	87.79%	–
Markov random field CNN [47]	95.44	99.75
Parallelized CNN [48]	–	99.62%
EDEN [25]	–	99.30%
Committee of 7 CNNs [35]	92.42%	99.19%