Training—Learning Through Gradient Descent
Explore how to train neural networks by applying gradient descent to update weights and reduce error. Understand why perceptron training methods need adjustment for multilayer networks and how calculus helps optimize learning.
Training a neural network
The central question of this lesson is: When we go from a single neuron (perceptron) to a multilayer neural network, would the training algorithm that we used for the perceptron also need to change?
Let’s take stock of the situation
It’s always a good idea to take a step back and take stock of the things we know. Things that we know will still not change when we complicate the model from a perceptron to a neural network:
A neural network still has the same two inputs: acting (0-10) and direction (0-10).
It still has one output (Good movie = 1, Bad movie = 0).
Each neuron in the neural network is still performing the same operation as that performed by the single neuron in a perceptron:
...