Neural Networks that can approximate any functions given a sample of data points. These Neural Network uses Conjugate Gradient as an update method and is written in C. Must include the INPUT.txt in the same folder as the .cpp file.
This code was used to generate the data in this research paper: https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.119.150601#fulltext
Last change since August 2016
Must include the following:
trainfile - training file name
OPTION - which method of training (mentioned more below)
ITERATIONS - number of iterations to run
SAVE - save progress every mentioned iterations
CUTOFF - number of maximum conjugate gradients to compute before ending iteration
Possible OPTIONs:
-2 = training network with old weights + noise
-1 = training network with old weights
0 = output cost function of data with old weights
1 = training network with new weights
To add old weights, follow the output weight's format. An example is shown in the sample INPUT.txt.
A Neural Network that approximates a function space based a sample of the gradients of the function.
A Neural Network that approximates a function space based a sample of the function.