Would it be possible to make the hyperparameters optimizable? Maybe it's the biologist in me, but I can't really see a reason why you couldn't just let the algorithm decide for itself which values of those parameters would work best.
Maybe I wasn't clear, or misunderstood some of the article, but I thought the article was about optimising the hyperparameters before the machine learning part happens. I meant optimising the parameters during the machine learning iterations.
