[机器学习]Lecture 1：Deep Learning，Backpropagation

目录

Introduction of Deep Learning

  Three Steps for Deep Learning:

define a set of function

goodness of function

pick the best function

Backpropagation

Gradient Descent：

Backpropagation——Forward pass and Backward pass

---

Introduction of Deep Learning

  Three Steps for Deep Learning:

• define a set of function

        决定一个好的function set很关键，即决定network的structure。

        Fully Connect Feedforward Network

        Deep = Many Hidden Layers

        Neural Network运算过程如下图，整个Neural Network的运算就是一连串matrix的运算：

• goodness of function

1. Find a function in function set that minimizes total loss L.
2. Find the network parameters θ* that minimize total loss L——Gradient Descent. Gradient Descent过程如下，找到一组好的参数后neural network的training就完成了

• pick the best function

Backpropagation

——如果要用Gradient Descent的方法train一个neural network，应该怎么做。

• Gradient Descent：

        Backpropagation不是一个和Gradient Descent不同的training的方法，它是Gradient Descent的一个比较有效率的演算法。

• Backpropagation——Forward pass and Backward pass

李老师视频中有详细讲解Forward pass和Backward pass的计算过程，听视频再复习一下，链接：

https://www.youtube.com/watch?v=ibJpTrp5mcE