• cross entropy loss is not quite the same as optimizing classification accuracy. Althougth the two are correlated.
  • It's not necessarily true that Deep learning approaches are often said to require enormous amount of data to work well. In this competitation, there'are 30,000 examples for 121 classes.
    To achieve this, some tricks are :

    • dropout
    • weight decay
    • data argumentation
    • pre-training
    • pseudo-labeling
    • parameter sharing
  • The method is implemented based on Theano:
    • Python, Numpy, Theano, cuDNN, PyCUDA, Lasagne
    • scikit-image: pre-processing and data argumentation
    • ghalton: quasi-random number generation
  • Hardware:
    • GTX 980, GTX 680, Tesla K40
  • Pre-processing and data argumentation:
    • Normalization: pre-pixel zero mean unit variance

to be finished

"Classifying plankton with deep neural networks" notes的更多相关文章

  1. Classifying plankton with deep neural networks

    Classifying plankton with deep neural networks The National Data Science Bowl, a data science compet ...

  2. Training Deep Neural Networks

    http://handong1587.github.io/deep_learning/2015/10/09/training-dnn.html  //转载于 Training Deep Neural ...

  3. 课程一(Neural Networks and Deep Learning)，第四周（Deep Neural Networks） —— 3.Programming Assignments: Deep Neural Network - Application

    Deep Neural Network - Application Congratulations! Welcome to the fourth programming exercise of the ...

  4. Training (deep) Neural Networks Part: 1

    Training (deep) Neural Networks Part: 1 Nowadays training deep learning models have become extremely ...

  5. 为什么深度神经网络难以训练Why are deep neural networks hard to train?

    Imagine you're an engineer who has been asked to design a computer from scratch. One day you're work ...

  6. [C4] Andrew Ng - Improving Deep Neural Networks: Hyperparameter tuning, Regularization and Optimization

    About this Course This course will teach you the "magic" of getting deep learning to work ...

  7. On Explainability of Deep Neural Networks

    On Explainability of Deep Neural Networks « Learning F# Functional Data Structures and Algorithms is ...

  8. Introduction to Deep Neural Networks

    Introduction to Deep Neural Networks Neural networks are a set of algorithms, modeled loosely after ...

  9. 深度神经网络入门教程Deep Neural Networks: A Getting Started Tutorial

    Deep Neural Networks are the more computationally powerful cousins to regular neural networks. Learn ...

随机推荐

  1. 关于OC中的几种代码延迟执行方式

    第一种: [UIView animateWithDuration:3 delay:3 options:1 animations:^{         self.btn.transform = CGAf ...

  2. TCP的拥塞控制（转载）

    1.引言 计算机网络中的带宽.交换结点中的缓存和处理机等,都是网络的资源.在某段时间,若对网络中某一资源的需求超过了该资源所能提供的可用部分,网络的性能就会变坏.这种情况就叫做拥塞. 拥塞控制就是防止 ...

  3. java二维码之利用谷歌的zxing生成二维码，解析二维码

    生成二维码 @RequestMapping("/123") public void test(HttpServletRequest request,HttpServletRespo ...

  4. RtlInitUnicodeString、IoCreateDevice、IoCreateSymbolicLink、IoDeleteDevice 四个 API 驱动函数的使用

    要解释"驱动对象",就得先从 DriverEntry() 说起: 做过C语言开发的都知道程序是从 main() 函数开始执行.在进行 Windows 驱动程序开发的时候没有 mai ...

  5. Java Calendar获取年、月、日、时间

    Java Calendar获取年.月.日.时间 Calendar c = Calendar.getInstance(TimeZone.getTimeZone("GMT+08:00" ...

  6. CodeForces 214B Hometask

    本题求n个数组成的大数,要求是2,3,5的倍数. 因为是2 和5 的倍数,所以个位为 0:所以若n个数中没有0,直接输出-1: 难点就是要求为3 的倍数. 因为若某个数为3的倍数,则其各位数的和必然是 ...

  7. 合成孔径雷达(Synthetic Aperture Radar, SAR)

    合成孔径雷达(Synthetic Aperture Radar, SAR)是20世纪最先进的科技发明之一.SAR有机载与卫载系统之分,能提供地表地形.乃至行星等远距离目标区高解析度图像.目前无人飞行载 ...

  8. kvm-GT

    REF: http://los-vmm.sc.intel.com/wiki/How-to-setup-kvmgthttp://xenvgt.sh.intel.com/image/bdw-h/ Host ...

  9. asp.net调用非托管dll，无法加载 DLL，找不到指定模块解决方法。

    最近开发一个项目,里面用到了非.net开发的一个dll文件接口,发现发布到window2003服务器上后,运行网站总是提示 "无法加载 DLL"D:\11\1.dll": ...

  10. eclipse ctrl链接设置

    选择[Window]菜单 Preferences ——>General——>Editors——>Text Editors——>Hyperlinking