Module 14 2022

01/04/2022

Training the network

All of the weights and biases are randomly set to start with, and the network will have no predictive ability at all.

Once we have obtained our first set of (incorrect) predictions for all of our input data, a total error is calculated for the data set, and the value of the weights and balances are adjusted in the direction that will slightly reduce the error. This process then repeats.

The Organisation for Professionals in Regulatory Affairs

5

Training the network

Over many cycles, the total error of the network for the training data is reduced until reaching a local minimum (not necessarily a global minimum)

The Organisation for Professionals in Regulatory Affairs

6

3