Research Article
Two-Stage Approach for Protein Superfamily Classification
Let = number of training samples | = number of randomly chosen hidden centres. | = desired output matrix of size where is the number of nodes in the output layer. | = neural network output matrix of size . | = hidden layer output matrix of size | = error matrix of size . | = connecting weight between the hidden and output layer of size . | Step 1. Perform QR decomposition of matrix. | Step 2. Evaluate where is of size and is of size . | Step 3. Evaluate the loss function | Step 4. Evaluate the Akaike's final prediction error | Step 5. Remove each network center and compute the loss functions. | | { | for every hidden node to do | | | | | | | | end for | | | } | if then | | | | | | | else | Compute the final optimal weight matrix from the equation | end if |
|