Let W_ij represents weight between node i at layer k and node j at layer (k – 1) of a given multilayer perceptron. The weight updation using gradient descent method is given by

Where α and E represents learning rate and Error in the output respectively.

Concept:

A perceptron is a single neuron model that was part of larger neural network. A multilayer perceptron is a finite acyclic graph. Nodes are neurons with logistic activation.

Explanation:

During multi-layer perceptron, neurons of i^th layer serve as input features for neuron of i+1^th layer. Using gradient descent algorithm, weights are updated incrementally after each pass over training data set. Gradient descent method assists in determining error while searching for optimal value to plug into cost function. It involves two steps:

• Calculate gradients of loss/error function
• Then updating existing parameters in response to the gradients.

This learning process is described by :                        

\({W_{ij}}\left( {t + 1} \right) = {W_{ij}}\left( t \right) - \alpha \frac{{\partial E}}{{\partial {W_{ij}}}},\;0 \le \alpha \le 1\)