A collection of neural network (non-linear) regression demos implemented in Torch7.
Training is performed via stochastic gradient descent (SGD). Alternative modes of fitting the input include an evolutionary strategy and random search by resampling weights and biases from the standard gaussian distribution.