Description
A C library for building neural networks with the capability to define custom activation functions, learning rate, bias and others parameters.
Both forward and backpropagation are built to parallelize through threads the operations.
In forward propagation each layer execute the activation functions in concurrent threads, when one layer end its own job, the results to the each next layer's neuron are collected with others concurrent threads. The whole process is executed from each layer from the input one to the output one.
In backward propagation first there is the delta evaluation from the output layer to the input layer, after that there is the weight updating process from the input to the output. Each return to the previous layer and delta evaluation on neurons in the same layer, is built, even in this case, with concurrent threads.
Due to the list type structure of layers and neurons, N3 Library allow you to remove and add neurons or layers even while it's in running mode ( at the end of each iteration ).