Deep Learning: Assuming a deep neural network is properly regulated, can adding more layers actually make the performance degrade?
2015-08-04 19:55
471 查看
Deep Learning: Assuming a deep neural network is properly regulated, can adding more layers actually make the performance degrade?
I found this to be really puzzling. A deeper NN is supposed to be more powerful or at least equal to a shallower NN. I have already used dropout to prevent overfitting. How can the performance be degraded?
Re-Ask
Follow17
Yoshua's Answer
View 2 Other Answers
![](http://qph.is.quoracdn.net/main-thumb-1090487-50-pbojqbgizyyymsvbijnvjurriryhegdd.jpeg)
Yoshua Bengio, My lab has been one of the three that started the deep learning approach, bac...
Upvoted by Prateek Tandon, Robotics and Strong Artificial Intelligence Researcher• Paul King, Computational Neuroscientist, Technology Entrepreneur • Jack Rae,Google DeepMind Research Engineer
Yoshua has 25 endorsements in Deep Learning.
If you do not change the size of the layers and just add more layers, capacity should increase, so you could be overfitting. However, you should check whether training error increases or decreases. If it increases (which is also very plausible), it means that adding the layer made the optimization harder, with the optimization methods and initialization that you are using. That could also explain your problem. However, if training error decreases and test error increases, you are overfitting.
I found this to be really puzzling. A deeper NN is supposed to be more powerful or at least equal to a shallower NN. I have already used dropout to prevent overfitting. How can the performance be degraded?
Re-Ask
Follow17
Yoshua's Answer
View 2 Other Answers
![](http://qph.is.quoracdn.net/main-thumb-1090487-50-pbojqbgizyyymsvbijnvjurriryhegdd.jpeg)
Yoshua Bengio, My lab has been one of the three that started the deep learning approach, bac...
Upvoted by Prateek Tandon, Robotics and Strong Artificial Intelligence Researcher• Paul King, Computational Neuroscientist, Technology Entrepreneur • Jack Rae,Google DeepMind Research Engineer
Yoshua has 25 endorsements in Deep Learning.
If you do not change the size of the layers and just add more layers, capacity should increase, so you could be overfitting. However, you should check whether training error increases or decreases. If it increases (which is also very plausible), it means that adding the layer made the optimization harder, with the optimization methods and initialization that you are using. That could also explain your problem. However, if training error decreases and test error increases, you are overfitting.
相关文章推荐
- 阅读linux内核需要的名词注释
- hdu5348(2015多校5)--MZL's endless loop(搜索)
- opencv 2.4.11 cuda5.5 总结
- 用 VIPER 构建 iOS 应用架构(1)
- 半自动化运维之服务器信息维护
- [MetaHook] Quake OpenGL function
- Linux C语言程序设计(十)——makefile
- linux 把文件安装到指定路径
- [MetaHook] Load DTX texture to OpenGL
- 用 VIPER 构建 iOS 应用架构(1)
- linux echo命令的-n、-e两个参数
- ORA-27041: unable to open file--恢复被rm意外删除数据文件
- VMware centos 设置上网的步骤
- linux回调函数
- 在Linux下安装和使用MySQL
- hdu 5349 维护序列 OR Treap
- 使用 jsch 连接linux
- shell知识点小结
- 微信公众号开发之被逼学web服务端1-----使用SecureCRT连接Linux服务器
- Linux下tomcat作为守护进程运行(开机启动、以指定的用户运行、解决非root身份不能绑定1024以下端口的问题)的配置方法