Failure is the discrepancy between a neural network’s actual output and the expected output. During backpropagation, the network measures this error after a signal passes through the network, then distributes corrections toward the neurons most responsible. section 7.2
Backpropagation treats failure as a correction signal:
- Calculate an output neuron’s delta from its error. section 7.2
- Propagate deltas backward through hidden layers. section 7.2
- Update weights using prior weights, prior inputs, deltas, and a user-determined learning rate. section 7.2
Figure 7.4. The mechanism by which an output neuron’s delta is calculated during the backpropagation phase of training
Shows how an output neuron’s delta is calculated during backpropagation. section 7.2
Figure 7.6. The weights of every hidden layer and output layer neuron are updated using the deltas calculated in the previous steps, the prior weights, the prior inputs, and a user-determined learning rate.
Shows how calculated deltas drive weight updates across hidden and output layers. section 7.2
