2.1 Principles of nerve conduction
y=activation(x*w+b)
Activation functions are usually nonlinear functions Sigmoid function and ReLU function
2.2 Simulate neural network with matrix operation
y=activation(x*w+b)
output = activation function (input * weight + bias)
2.3 Multilayer Perceptron Model
1 Recognition of minst handwritten digit images with a multilayer perceptron model
The data of the input layer is a 28*28 two-dimensional image converted to a 1-dimensional vector by reshape as a shuru of 784 neurons
The input layer has 784 input neurons to receive external signals
The hidden layer simulates internal neurons with a total of 256 hidden neurons
The 10 output neurons in the output layer are the predicted results
There are 10 results corresponding to the numbers 0-9 we want to predict