Design parameters as output ML
I feel there is no need to predict DPs when they are part of the outputML, as this is an ill-posed task, given that the DPs are just sampled randomly and then in principle it should be possible to estimate DP5 from the rest. Therefore, we need to check the results of this specific case, to see if the DP in the outputML can be predicted.
As a solution, we can just add the DP both the Input and Output, such that the Network learns the almost direct relation, and then during generation is almost fixed directly in the generated InputML