Simple implementation of feedforward neural network BPNN

Divided into 4 steps

initialization

Use random functions to assign weights and biases. In addition to the input layer, biases need to be added.

Calculate output

Calculate the output of each neuron

Calculation error

Error calculation of output layer

Error calculation of intermediate layer

Adjust weights and biases

Among them, L is the learning rate, which can be set to 0.9. Next, adjust the weight

accomplish

The matlab implementation code is as follows

%% 简单前馈神经网络

% 初始化权重和偏置
W12=rand([3,2]);
W23=rand([2,1]);

B12=rand([1,2]);
B23=rand([1,1]);
% 输出
input=[1,0,1];

for i=1:1000
    % 计算输出
    S2=input*W12+B12;
    O2=(1+exp(-S2)).^-1;
    S3=O2*W23+B23;
    O3=(1+exp(-S3)).^-1;

    % 计算误差
    E3=O3.*(1-O3).*(1-O3);
    E2=O2.*(1-O2).*(sum(W23*E3,2).'); 

    % 调整
    W23=W23+(0.9)*O3.*E3;
    W12=W12+(0.9)*O2.*E2;

    B12=B12+(0.9)*E2;
    B23=B23+(0.9)*E3;
end