梯度下降算法

1.神经网络基本算法—梯度下降算法

每一条线代表一个权重，和输入值相乘，然后相加==z z/1==隐藏层




perceptrons 表示0 1之间取值的神经网络层



输入层 每个点都是一个像素点* 权重

 首先数组分层，将数字分开<br />    <br />     ![image.png](https://cdn.nlark.com/yuque/0/2021/png/1267880/1622304977064-f55dfe0a-bf70-4ee7-98b1-41026b6fd816.png#align=left&display=inline&height=241&margin=%5Bobject%20Object%5D&name=image.png&originHeight=481&originWidth=659&size=235368&status=done&style=none&width=329.5)<br />    
   像素值 0-255 转换为 0-1之前的值（x/255）,然后进行计算
  <br />![image.png](https://cdn.nlark.com/yuque/0/2021/png/1267880/1622304981861-4aedc097-56dc-4ef4-b488-bcfb8c207e3a.png#align=left&display=inline&height=158&margin=%5Bobject%20Object%5D&name=image.png&originHeight=315&originWidth=594&size=111500&status=done&style=none&width=297)<br />![image.png](https://cdn.nlark.com/yuque/0/2021/png/1267880/1622304994392-c47ac37e-4dd0-497e-b730-2e6e057cb9e4.png#align=left&display=inline&height=276&margin=%5Bobject%20Object%5D&name=image.png&originHeight=552&originWidth=637&size=226244&status=done&style=none&width=318.5)<br />      <br />        <br />      C 损失函数 或者目标函数<br />         <br />    ![image.png](https://cdn.nlark.com/yuque/0/2021/png/1267880/1622305158137-d9d789ef-f1aa-4e8b-87fb-b04fd15df405.png#align=left&display=inline&height=168&margin=%5Bobject%20Object%5D&name=image.png&originHeight=335&originWidth=559&size=107316&status=done&style=none&width=279.5)![image.png](https://cdn.nlark.com/yuque/0/2021/png/1267880/1622305003492-3934524b-42e4-4128-b096-d7213d556908.png#align=left&display=inline&height=168&margin=%5Bobject%20Object%5D&name=image.png&originHeight=335&originWidth=559&size=107316&status=done&style=none&width=279.5)
![image.png](https://cdn.nlark.com/yuque/0/2021/png/1267880/1622305007856-83f6b13e-41a8-43df-86a8-20cf59e9a94d.png#align=left&display=inline&height=220&margin=%5Bobject%20Object%5D&name=image.png&originHeight=439&originWidth=618&size=171929&status=done&style=none&width=309)<br />![image.png](https://cdn.nlark.com/yuque/0/2021/png/1267880/1622305012217-8beeed94-2488-4ea2-a7a7-5e90498574db.png#align=left&display=inline&height=137&margin=%5Bobject%20Object%5D&name=image.png&originHeight=274&originWidth=443&size=69556&status=done&style=none&width=221.5)<br />     <br />       曲线 切线 斜率 ？？<br />       求导数 ==切线<br />       斜率 =tangent
   ![image.png](https://cdn.nlark.com/yuque/0/2021/png/1267880/1622305017924-16dc7e80-c678-4a33-8e5a-3ebfba81dad3.png#align=left&display=inline&height=224&margin=%5Bobject%20Object%5D&name=image.png&originHeight=447&originWidth=576&size=235908&status=done&style=none&width=288)<br />       <br />         n 学习率*偏导函数<br />       

2.梯度下降算法-变种

T代表转至 德尔塔C 变化量

3.梯度下降实现-上

<br />      ![image.png](https://cdn.nlark.com/yuque/0/2021/png/1267880/1622305061126-edb9e6f8-5ea7-4e8d-9f37-4a448b78cf82.png#align=left&display=inline&height=242&margin=%5Bobject%20Object%5D&name=image.png&originHeight=483&originWidth=581&size=150600&status=done&style=none&width=290.5)<br />     ![image.png](https://cdn.nlark.com/yuque/0/2021/png/1267880/1622305066818-de256c48-795b-4360-a31e-d23d294434ee.png#align=left&display=inline&height=77&margin=%5Bobject%20Object%5D&name=image.png&originHeight=154&originWidth=568&size=89366&status=done&style=none&width=284)<br />           <br />            w= 权重向量  a=输入  b=biase 偏向<br />            <br />          以下  向前传递算法、 随机梯度下降算法<br />            <br />      ![image.png](https://cdn.nlark.com/yuque/0/2021/png/1267880/1622305077064-44dc5668-140a-4063-a7f1-a43a64916bc0.png#align=left&display=inline&height=196&margin=%5Bobject%20Object%5D&name=image.png&originHeight=391&originWidth=703&size=241234&status=done&style=none&width=351.5)<br />         <br />        ![image.png](https://cdn.nlark.com/yuque/0/2021/png/1267880/1622305089011-da5068e2-999d-48d1-bf77-6f74775de98d.png#align=left&display=inline&height=22&margin=%5Bobject%20Object%5D&name=image.png&originHeight=43&originWidth=425&size=14848&status=done&style=none&width=212.5)<br />![image.png](https://cdn.nlark.com/yuque/0/2021/png/1267880/1622305094211-6513ce7a-5e15-47f4-901d-676312a1b280.png#align=left&display=inline&height=278&margin=%5Bobject%20Object%5D&name=image.png&originHeight=556&originWidth=735&size=338771&status=done&style=none&width=367.5)

3.梯度下降实现-下