Stanford ML - Lecture 5 - Neural Networks: Learning
来源:互联网 发布:芭芭拉史翠珊 知乎 编辑:程序博客网 时间:2024/04/28 00:55
1. Cost function
- Neural Network (Classification)
- Binary classification
- 1 output unit
- Multi-class classification (K classes)
- K output units
- Cost function
- Logistic regression:
- Neural network:
2. Backpropagation algorithm
- Gradient descent
- need code to compute
- need code to compute
Is it ??
- Gradient computation
然后根据backpropagation算法进行梯度的计算,这里引入了error变量δ,用来表示真实值与forward propagation计算值之间的差,也是梯度的主要依据来源。
我们定义神经网络的总误差为:而对于前面的一层(如第三层),其误差可以定义为:
分别代入即得
但是始终是不变的。
- Stanford ML - Lecture 5 - Neural Networks: Learning
- Stanford ML - Lecture 4 - Neural Networks: Representation
- Stanford ML - Neural Networks: Learning 神经网络训练学习
- Stanford ex4:Neural Networks Learning
- Stanford ML - Neural Networks: Representation 神经网络表示
- [ML of Andrew Ng]Week 5 Neural Networks:Learning
- Stanford 机器学习-Neural Networks learning
- Neural Networks: Learning(Andrew ng ML)
- Stanford ML - Lecture 6 - Advice for applying machine learning
- Stanford ML - Lecture 7 - Machine learning system design
- Stanford ML - Lecture 11 - Large scale machine learning
- 机器学习中的神经网络Neural Networks for Machine Learning:Lecture 5 Quiz
- Stanford机器学习---第5讲. 神经网络的学习 Neural Networks learning
- Stanford机器学习---神经网络的学习 Neural Networks learning
- Stanford 机器学习笔记 Week5 Neural Networks: Learning
- Stanford 机器学习 Week5 作业: Neural Networks: Learning
- 【Stanford机器学习笔记】7-Neural Networks: Learning
- Stanford ML - Lecture 9 - Clustering
- Mysql workbench中PK.NN等的意思
- SDUT 第九周周赛
- 黑马程序员-数组
- 几个OJ链接
- Judging Olympia
- Stanford ML - Lecture 5 - Neural Networks: Learning
- 关于计算机体系结构之RAM
- 第一个windows8应用程序
- PHP安装eAccelerator加速器
- 如何在C/C++中动态分配二维数组
- Spring aop拦截器
- JAVA基础知识
- java中String Date Timestamp Calendar 之间的关系及转换
- com使用自定义数据类型(UDT) [转]