![Hands-On Mathematics for Deep Learning](https://wfqqreader-1252317822.image.myqcloud.com/cover/81/36698081/b_36698081.jpg)
上QQ阅读APP看书,第一时间看更新
Chain rule
Let's take an arbitrary function f that takes variables x and y as input, and there is some change in either variable so that . Using this, we can find the change in f using the following:
![](https://epubservercos.yuewen.com/FF11E0/19470372701459106/epubprivate/OEBPS/Images/Chapter_1343.jpg?sign=1739637306-NYuZXdz8ZukzpmM1ynOdaC5Clw0wr7sw-0-35aec365b2c5cf56ce3700bfcb5e4d12)
This leads us to the following equation:
![](https://epubservercos.yuewen.com/FF11E0/19470372701459106/epubprivate/OEBPS/Images/Chapter_1002.jpg?sign=1739637306-Vhf1OnZEr8vjXk8ZxpdAUb5LyQKBPXBU-0-bb39114a39c2bdc4b898444927305b81)
Then, by taking the limit of the function as , we can derive the chain rule for partial derivatives.
We express this as follows:
![](https://epubservercos.yuewen.com/FF11E0/19470372701459106/epubprivate/OEBPS/Images/Chapter_1285.jpg?sign=1739637306-s2Ch5mcwmo5Ukitsd8k7KLStLG886vJd-0-e20537d2d30567a121922021a58a0434)
We now divide this equation by an additional small quantity (t) on which x and y are dependent, to find the gradient along . The preceding equation then becomes this one:
![](https://epubservercos.yuewen.com/FF11E0/19470372701459106/epubprivate/OEBPS/Images/Chapter_1007.jpg?sign=1739637306-T6QhDbZxBZN6OXXezcl7kUNBQYocDgpb-0-450126262bf904fb64d260c806130f8a)
The differentiation rules that we came across earlier still apply here and can be extended to the multivariable case.