This is not exactly an answer to your question, but more of a comment.

In what concerns the educational point of view, I came across these two sources of information that might interest you:

• Simulating Neural networks with Mathematica, by James Freeman, 1994, Addison-Wesley -> this is a very very old book. For sure that its content is out of date in respect to current best practices both in what concerns Mathematica and the ANN topic. But for learning purposes, probably you can find something interesting in it.

• PSY 5038W - Introduction to Neural Networks . I haven't read the material (yet...).

Of course, there's a lot of other stuff going on in what concerns ANN and the Mathematica world, but of much less interest to your learning question. Nevertheless, for the sake of completeness, I'm adding other initiatives, some well known, and others less talked about:

• for sure you are aware of all the work being developed by WRI: overview of machine learning in the WL & a shallow tour of deep learning

• you mentioned a package. I imagine that you were referring to Deep Learning Library for Mathematica

• CaffeLink & CaffeLink at IMS2015

• GPU-Accelerated Recurrent Neural Networks at IMS2015

I hope this helps you in some way, and please feel free to add to it,