/Neu

Neu is an auto-differention library to construct, train and test deep neural networks written in C++14.

Primary LanguageC++MIT LicenseMIT

Neu

Neu is an auto-differention library to construct, train and test deep neural networks. Neu is written in C++14.

This library is released under the MIT License, see LICENSE.

Learned first conv filters of example/cifar10

Learning curve of example/cifar10_deepcnet
(DeepCNet(5,300), the purple is train cross entropy error, the green is test one)

It took about 48hour with GTX980

Dependency

Features

Layer

  • inner production
  • spacial convolution
  • max pooling
  • average pooling
  • dropout
  • batch normalization

Activation

  • ReLU
  • leaky ReLU
  • sigmoid
  • tanh

Optimizer

  • fixed learning rate
  • momentum

Installation

mkdir build && cd build
cmake ..
make install