Table 1
From: Extreme learning machine versus classical feedforward network
Algorithm 1 Gradient Descent Method for neural network training | |
|---|---|
Input: \(\mathcal {D} = (\mathbf {X}, \mathbf {Y})\) - training data, | |
Output:\({\mathbf {w}}\) - final weights values | |
Require: \(\mathcal {L}\) - the cost function minimised during training, | |
\(\mathcal {D} = (\mathbf {X}, \mathbf {Y})\) - training data, | |
f - the transformation performed by a network trained with the weights \(\mathbf {w}\), | |
\(\alpha \) - learning coefficient | |
\(\mathbf {w} \leftarrow \) initial randomly assigned weights | |
1: repeat | |
2: \(\hat{\mathbf {Y}} \leftarrow f(\mathbf {X}, \mathbf {w})\) | |
3: \(\varDelta {\mathbf {w}} \leftarrow \frac{\partial \mathcal {L}(\hat{\mathbf {Y}}, \mathbf {Y})}{\partial \mathbf {w}}\) | |
4: \(\mathbf {w} \leftarrow \mathbf {w} - \alpha \varDelta {\mathbf {w}}\) | |
5: until stopping criterion is false | |
6: return \(\mathbf {w}\) |