On Training Efficiency and Computational Costs of a Feed Forward Neural Network: A Review

Table 6

Summary of adaptive strategies.

Ref.	Method	Convergence	Precision	Computational costs	Notes
[34]	Scalable sigmoid; (NNAAF-1)	6960 Epochs	% error = 0.033	Training time: 2739 s	Learning rate 0.9
[34]	Sin-sigmoid; (NNAAF-2)	8232 Epochs	% error = 0.045	Training time: 2080 s	Learning rate 0.9
[34]	Morlet wavelet; (NNAAF-3)	10000 Epochs	% error = 0.097	Training time: 3046 s	Learning rate 0.2
[35]	Sigmoid-radial-sin	5250 Epochs	N-RMSE = 0.09301	N/A	Trained with Levenberg Marquardt Algorithm
[36]	Sin-sigmoid	5000–9000 Epochs	89.60–94.3% accuracy (classification problem)	N/A	Implemented on higher order NN (HONN)
[37]	Trainable AF	20000 Epochs	RMSE = −35 dB	N/A