MATLAB Feed Forward Neural Networks with Back Propagation
The implementations provided here do not require any toolboxes, especially no neural network toolbox.
The neural network implementations in this repo are set up in three complexities:
simpleTrainExample: Script which trains a neural network of 3 layers (in, hidden, out), each consisting of only a single neuron. Not really functional for any task, primarily available for understanding the basic behavior.
trainNeuralNetStatic: Function which is capable of training a basic neural network of 3 layers with an arbitrary amount of neurons each. Look at this before looking at the next function! Takes a network generated by the generateNeuralNet function and uses applyNeuralNet.
trainNeuralNet : Function which trains networks of arbitrary size. Basically wraps the backpropagation behavior of trainNeuralNetStatic into an extra loop in order to handle more layers.
请发表评论