How is Pareto mode used in applying Tikhonov regularization constraints?
Regularization constitutes a trade-off situation. Where there is insufficient data to estimate all model parameters uniquely (which is always the case given the innate complexity of reality) the shortfall in information must be made up by the modeller. That is not to say that he/she can provide correct (or even nearly correct) values for parameters when making up for this information shortfall. But he/she can provide values for parameters, or for relationships between parameters, that are of “minimum error variance” when all of his/her expertise is taken into account. Note that “minimum error variance” does not mean “small error variance”. It only means that the potential for being wrong is minimized because that potential has been made symmetrical with respect to the estimates provided. Thus there is as much chance of being wrong one way as there is of being wrong the other way. So when we calibrate a highly parameterized model, we supply a backup plan for all parameters. In doing thi
