Make Your Own Neural Network

前言：这篇文章介绍一本小书，《Make your own neural network》，书籍讲解了神经网络的原理以及使用python实现神经网络。文章记录核心知识，对书中的公式再进行推导。整体而言书籍讲解比较通俗，不足之处在于体系比较散、没有说清反向传播与梯度下降的关系。

Chapter 1 How They Work

英里与千米的转化

Again withoutgetting toodistracted by exactwaysof workingout c，and to remainfocussedon thisidea of successivelyrefining it,we could suggest thatthe correctionis a fractionof the error. That'sintuitivelyright - a big errormeans a bigger correction is needed,and a tiny error means we need theteeniest ofnudges to c.

kilometres convert to miles

Key Points:

• All useful computer systems have an inpit, and an output,with some kind ofcalculation in between. Neyralnetworks are no different.
• When we don'tknow exactly how something works we can try to estimate it with a model which includesparameters which we can adjust. If we didn't knowhow to convert kilometres to miles, we might use a linearfunction as a model, withan adjustable gradient.
• A good way ofrefining these models is to adjustthe parametersbased on how wrongthe model is compared to known true examples.

训练简单的分类器

三层神经网络

A、前向传播计算

B、反向传播解释

C、权重更新

引：梯度下降是一种优化算法，为了求损失函数的极值。反向传播算法是一种更适用于多层神经网络结构特的梯度下降法（使用反向传播算法更容易求解梯度）。

总结：

整体下来书籍阅读比较顺畅，有些需要计算的地方值得停下来耐心算一算。书籍第一章围绕神经网络的理论进行讲解，通过简单的千米与英里转化作为例子导入，以毛虫与瓢虫的分类器讲解前向传播，而后通过具体讲解三层神经网络的反向传播算法，讲清楚了神经网络训练的具体原理。

第二章围绕使用Python如何构建神经网络进行阐述，书籍中也包含源码。此外还发现一些值得参考的视频，也简单整理如下：

• http://makeyourownneuralnetwork.blogspot.com/
• https://www.bilibili.com/video/BV1aB4y13761/?spm_id_from=333.337.search-card.all.click&vd_source=1542a52f04335c1d8415d392ab399b7e