AbdulFMS / cnn_from_scratch

A simple implementation of a convolutional neural network from scratch in C++

Neural Network in Pure C++

Simple modular implementation of a neural network in C++ using only the STL.

Installation

Get the MNIST data set:

bash get_mnist.sh

Generate your Makefile:

cmake -DCMAKE_BUILD_TYPE=Release

Make the code:

make

Run:

./neural_net_in_cpp data

The training should take about a minute and achieve ~97% accuracy.

Todos

• Fully connected;
• Sigmoid;
• Dropout;
• ReLU;
• Tanh;
• Leaky ReLU;
• Batch normalization;
• Convolutional layers;
• Max pooling;
• Other optimizers (Adam, RMSProp, etc);
• Learning rate scheduler;
• Plots;
• Filter visualization
• CUDA?

License

MIT