Improved neural network Monte Carlo simulation

The algorithm for Monte Carlo simulation of parton-level events based on an Artificial Neural Network (ANN) proposed in Ref.~ is used to perform a H\to 4\ell H?4? decay. Improvements the training have been implemented avoid numerical instabilities. integrated decay width evaluated by ANN within 0.7% true value and unweighting efficiency 26% reached. While not automatically bijective between input output spaces, which can lead issues with quality, we argue that procedure naturally prefers maps, demonstrate trained very good approximation.