CS231n Neural Networks Part 3: Learning and Evaluation Babysitting the learning process
来源:互联网 发布:js 时间间隔 编辑:程序博客网 时间:2024/05/22 15:22
1.Loss function
如果是mini-batch,那么loss一般是抖动下降的,对于不同的lr有着不同的loss下降的曲线。
2.Train/Val accuracy
不同的acc曲线对应着不同的含义,比如说一般来说epoch上升,acc也会上升,但是如果过拟合了,acc在val会下降。
3.Ratio of weights:updates
如果将grad和weights变成向量,那么两个向量的长度的比例可以看作是ratio,其中ratio是1e-3左右。
4.Activation/Gradient distributions per layer
每一层的max min 直方图分布等,convNetJs上面有demo
5.First-layer Visualizations
可以看第一层的特征,第一层一般是纹理特征,如果学习正确,会是一些线条来提取线条的分布用的。
0 0
- CS231n Neural Networks Part 3: Learning and Evaluation Babysitting the learning process
- CS231n Neural Networks Part 3: Learning and Evaluation Gradient Checks
- 20170307#cs231n#7.Neural Networks Part 3: Learning and Evaluation
- CS231n Neural Networks Part 3: Learning and Evaluation sanity checks Tips/Tricks
- Babysitting the Learning Process
- Neural Networks and Deep Learning
- Neural Networks and Deep Learning
- Neural networks and Deep Learning
- Neural Networks and Deep Learning
- Machine Learning - Neural Networks Representation Part I
- Machine Learning - Neural Networks Representation Part II
- A Tutorial on Deep Learning Part 2: Autoencoders, Convolutional Neural Networks and Recurrent Neural
- CS231n Neural Networks Part 2: Setting up the Data and the Loss
- 20170304#cs231n#6.Neural Networks Part 2: Setting up the Data and the Loss
- 3天完成Neural Networks and Deep learning课程
- BOOK: Neural Networks and Learning Machines
- 《Neural networks and deep learning》概览
- 《Neural networks and deep learning》概览
- Java 类和对象的初始化过程及初始化顺序
- LeetCode | Find Minimum in Rotated Sorted Array
- poj 1195 Mobile phones
- HTML5与CSS3基础教程-笔记
- JSON数组形式字符串转换为Map数组(转为其他的Bean的话,请参考自行变形)
- CS231n Neural Networks Part 3: Learning and Evaluation Babysitting the learning process
- MFC---利用标题对列表进行排序
- 欢迎使用CSDN-markdown编辑器
- Linux命令 --- 文件操作篇
- LintCode:子集
- app上线需要做的测试
- 嵌入式linux 内核和根文件系统烧写方式简介
- Pyhton日记——Flask写一个简易的知乎日报API
- PAT乙级.1009. 说反话 (20)