我创建了一个Octave脚本,用于使用反向传播训练具有1个隐藏层的神经网络,但它似乎不适合XOR函数.
x 输入4x2矩阵 [0 0; 0 1; 1 0; 1 1]y 输出4x1矩阵 [0; 1; 1; 0]theta 隐藏/输出图层权重z 加权总和a 激活函数应用于加权和m样品数量(4这里)我的权重初始化如下
epsilon_init = 0.12;
theta1 = rand(hiddenCount, inputCount + 1) * 2 * epsilon_init * epsilon_init;
theta2 = rand(outputCount, hiddenCount + 1) * 2 * epsilon_init * epsilon_init;
Run Code Online (Sandbox Code Playgroud)
前馈
a1 = x;
a1_with_bias = [ones(m, 1) a1];
z2 = a1_with_bias * theta1';
a2 = sigmoid(z2);
a2_with_bias = [ones(size(a2, 1), 1) a2];
z3 = …Run Code Online (Sandbox Code Playgroud) matlab machine-learning octave backpropagation neural-network