Because REVAC is implemented as a sequence of distributions with slowly decreasing Shannon entropy we can use the Shannon entropy of these distributions to estimate the minimum amount of information needed to reach a target performance level. We can also measure how this information is distributed over the parameters, resulting in a straightforward measure for parameter relevance. This measure can be used in several ways. First, it can be used to choose between different operators. An operator that needs little information to be tuned is more fault tolerant in the implementation, easier to calibrate and robuster against changes to the problem definition.
Second, it can be used to identify the critical parts of an EA. For this we calculate relevance as the normalized entropy, i.e., the fraction of total information invested in the particular parameter. When an EA needs to be adapted from one problem to another, relevant parameters need the most attention. With this knowledge, the practitioner can concentrate on the critical components straight away. Third, it can be used to define confidence intervals for parameter choices. Given a distribution that peaks out in a region of high probability (except for the early stage of the algorithms the marginal distributions have only one peak), we give the 25th and the 75th percentile of the distribution as a confidence interval for the parameter. That is, every value from this range leads to a high expected performance, under the condition that the other parameters are also chosen from their respective confidence interval.
