In this seminar I will present recent work on the use of statistical learning tools for the analysis of stochastic dynamic model of complex systems, biological ones in primis. Statistical learning offers powerful tools to manage uncertainty, particularly in model parameters. Statistical learning methods, combined with model simulations, can be used to estimate efficiently the probability of observing a given behaviour as a function of model parameters. This can be turned into a jackknife method to perform several tasks: optimise likelihoods efficiently, doing parameter synthesis, design, and control. At the same time, statistical learning provides novel ways to look into model abstraction, something we called statistical correction maps.  In this talk, I will give an overview of these ideas, focussing on some aspects with more detail.