NetMaker

Krzysztof M. Graczyk, Piotr Plonski, Robert Sulej

Neural Network Parameterizations of Electromagnetic Nucleon Form Factors

Parameterizations of Electromagnetic Nucleon Form Factors - numerical results

Bayesian Framework - go back to the method description

GraNet - neural network library in C++ by K. M. Graczyk

paper online on arXiv.org
paper online on JHEP

Best fit, in terms of the maximal evidence, is shown for each form factor. The fit uncertainties shown here has been computed using the Hessian approximation formula. The exact and finite differences formulas can be applied to calculations of fit uncertainties with the corresponding project files. Results for each form factor are available in a separate zip archive. Included files are following:

gXX_data.dat - ascii file, experimental data with uncertainties (constraint point is included at the file end), each row consists of: Q², G(Q²), σ_G(Q²);
gXX_fit.dat - fit numerical results, file format is like for gXX_data.dat;

gXX.NetAscii - neural network description in NetMaker format (XML file with the network weights, the optimal regularization factor value α used for training, the log evidence, and the network error on training data χ²/N);

gXX.NetPrj - NetMaker project set up for calculating the fit values from the included network file or recreating the network from scratch.