Apache Spark Deep Learning Cookbook

How it works...

To train feedforward neural networks the most commonly used technique is backpropagation through time. It is a supervised learning method used to reduce the loss function by updating weights and biases in the network after every time step. A number of training cycles (also known as epochs) are executed where the error determined by the loss function is backward propagated by a technique called gradient descent. At the end of each training cycle, the network updates its weights and biases to produce an output which is closer to the desired output, until a sufficiently small error is achieved :

1. The backpropagation algorithm basically implements the following three fundamental steps during every iteration:
   • The forward pass ...