NobleNeuron

About

A library in C++/Armadillo and Fortran for simulating neural networks. Currently the C++ version supports multilayer feedforward and convolutional networks.

File Structure

• FeedForward.cpp (Create a feedforward multilayer network: add layer, feedforward, backpropagate)
  • ff_layer.cpp (Execute the math of each layer of the feedforward network)
    • activ.cpp (Activation ande deactivation functions)
• Convolutional.cpp (Create a convolutional network: add layer, feedforward, backpropagate)
  • conv_layer.cpp (Execute de math of a convolutional layer)
    • activ.cpp
  • pooling_layer.cpp (Execute the math of a pooling layer)
• loss.coo (Loss functions)

Compile Noble Neuron

make
make install
make clean

Compile an Example

g++ main.cpp -o executable -lNobleNeural -larmadillo

Use Noble Neuron

FeedForward F(n), n is the size of the input vector

• F.add_layer(n,s), 'n' is the size of layer output and 's' is the activation;
• F.feedforward(M), 'M' is an input column vector;
• F.backpropagate(M), 'M' is the gradient of the loss function;
• F.applygrads(eta), Apply gradients, 'eta' is the learning rate;
• F.output(), Gives the output of the network;
• F.unflatten(rows,cols), Produces the gradient of the network as a rows x cols of a slice of a cube to be used in a convolutional network;

Convolutional C(height,width,depth), create a convolutional network with this input

• C.add_conv(fh,fw,channels,activation), Add a convolution layer with 'channels' fh x fw filters;
• C.add_down(fh,fw,sh,sw,kind), Add a down sampling layer. sh and sw are the stride, kind is the pooling = max or average;
• C.feedforward(D), 'D' is an input cube;
• C.backpropagate(D), 'D' is the gradient of the loss function;
• C.applygrads(eta)
• C.output()
• C.flatten(), Produces a column matrix to be used in a feedforward network;

Credits

@misc{daCunha2021,
    author       = {C. R. da Cunha},
    title        = {{NobleNeuron, a library for simulating neural networks.}},
    month        = apr,
    year         = 2021,
    version      = {1.0},
    publisher    = {GitHub},
    url          = {https://github.com/StxGuy/NobleNeuron}
    }

License

Copyright (c) 2021-2022 - Carlo R. da Cunha, "NobleNeuron"
carlo.requiao@gmail.com