Feed-Forward Neural Networks
The feed forward neural networks are also termed as the multi-layer perceptrons. In this type of neural networks, the observed movement of data is unidirectional. It may consist of either zero, one or many hidden layers. When we use them, they are fast in their operation. But while training the data, they may take a while. The nodes of this network do not form a cycle. Moreover, all the nodes in a particular network are connected with the nodes of the next layer. we may say that they are fully connected. There are no back-loops present. In order to minimise the error in the prediction, we employ the back-propagation algorithm so as to update the weights and biases.
The back-propagation algorithm focuses on minimising the error rate between the predicted output and the actual result by means of adjusting the values of the weights and biases. The final weights that help it provide the optimal result are regarded to be the solution. Thus, we effectively train our model using the chain rule. After each forward pass that is implemented, we perform a backward pass that adjusts the parameters.
Applications of Feed-forward Neural Networks
- Computer Vision: With a good number of training examples available, the hierarchical feed-forward neural networks are able to form the feature learning tool to build the visual recognition models that can be implemented across various domains.
- Speech Recognition: This is designed as a three layer system. The first layer analyzes and classifies a set of speech attributes that are assigned to it. The middle layer is responsible for building the speech matrix from the classification scores that have been developed by the first layer. Then comes the third or the last layer that evaluates this speech matrix so as to predict the output for the model.
- Handwritten Character Recognition: The training set provided to the model will be a bitmap pattern of the handwritten characters as the input to the model. The right letter or digit is the output that we desire.
