RNN 2：反馈神经网络的解决OXR问题

最新推荐文章于 2024-01-16 08:26:31 发布

yanmingtop

最新推荐文章于 2024-01-16 08:26:31 发布

阅读量1.2k

点赞数

分类专栏：机器学习文章标签：神经网络 RNN Recurrent

本文链接：https://blog.csdn.net/yanmingtop/article/details/40080579

版权

利用recurrent network 解决xor的分类问题

%% 3 units recurrent，2 circle delay
clc
clear 
close all

%% net config
f = @(x) 1./(1+exp(-x));
delta = @(i,j) i==j;
alpha = 4;
p1 = zeros(3,6);
p2 = zeros(3,6);
p3 = zeros(3,6);

%% train
x = randi([0,1],[2,1]);
y = zeros(3,1); % t = 0;
z = [y;x;1];
%w = -1 + 2 * rand(3, 6);  %权值为何要初始化为负值？
w = rand(3,6);
y = f(w*z);  % t=1
d = xor(x(1),x(2));
maxIter = 1000;
for n = 1:maxIter    
    x = randi([0,1],[2,1]);
    z = [y;x;1];
    y = f(w*z);  % t=2,3 .... n
    for i = 1:3 
        for j = 1:6
            q1 = y(1)*(1-y(1))*(w(1,1)*p1(i,j)+w(1,2)*p2(i,j)+w(1,3)*p3(i,j)+delta(1,i)*z(j));
            q2 = y(2)*(1-y(2))*(w(2,1)*p1(i,j)+w(2,2)*p2(i,j)+w(2,3)*p3(i,j)+delta(2,i)*z(j));
            q3 = y(3)*(1-y(3))*(w(3,1)*p1(i,j)+w(3,2)*p2(i,