Neural networks, probability distributions and structure functions

by Stefano FORTE (INFN Roma III)

Friday 30 Aug 2002, 14:00 → 15:00 Europe/Zurich

TH Auditorium (CERN)

I discuss some new ideas to deal with the treatment of the uncertainties on the QCD input to the computation of collider processes. I describe the extraction from the data of a nucleon structure function, together with errors and full point-to-point correlation, through the determination of the probability density in the space of structure functions. This is accomplished through the use of neural networks in a Monte Carlo approach. I then show how structure function data can be analyzed in a bias-free way by studying evolution equations for truncated moments of parton distributions. As an example of application of these techniques, I present a determination of the strong coupling alpha_S which minimizes theoretical uncertainties and bias.