I haven't read the paper yet but you present an interesting argument to think about but I am not sure that neural network based approaches are mutually exclusive to experimental design and can in fact inform them. I'd love to talk more about this.

Is anyone interested in a HN journal club?