Beginning Anomaly Detection Using

backpropagation to recalculate the 

weights and biases for each activated node.

In backpropagation, a 

cost function takes into account the model’s predictions 

for one pass of the training data through the network and what the actual predictions 

should be. The cost function gives you an indicator of how good the model’s weights are 

at predicting the correct outcome. For this example, assume that Figure 

3-11

 shows the 

formula of the cost function.

Figure 3-10.  Finally, the data from the second hidden layer goes to the output 

layer, where one neuron fires in this case

Chapter 3   IntroduCtIon to deep LearnIng

---

82

This cost function is called the