Understanding how neural networks learn
The prediction process shows that a neural network’s outputs are determined by the weights of its input connections. Adjusting these weights to better align with different input values improves the network’s accuracy and performance. This adjustment forms the core of the learning process, where the network iteratively updates its weights to minimize prediction errors. This iterative refinement of weights is known as training.
To train a neural network, the backpropagation algorithm is used to adjust the weights effectively. The process involves four key steps, which we’ll examine in the subsequent sections.
Step 1: Predicting outputs
In this step, a set of training data inputs is used by the neural network to predict the corresponding outputs. For instance, with reference to Figure 10.2, if the training data consists of the distance from the turret to the player and the firing angle error, the normalized inputs...
