Do machine learning / neural network questions go on scicomp or crossvalidated? Personally, I consider a question like https://scicomp.stackexchange.com/q/10248/3212 to be a better fit here than on crossvalidated, but the latter has many neural-network related question. Or is Compsci better if I want theoretical answers (how and why does it work), and Crossvalidated better if I want practical answers (how do I make it work)?
Given that some computational scientists use machine learning and neural network methods as part of their research, I'm inclined to say that these methods are on-topic for SciComp. (For example, these methods are being used to predict properties of biomass-derived fuel molecules given a training set and a list of functional groups.)
That said, you might get better answers at CrossValidated, since CrossValidated has more of a statistics and data science focus than SciComp.