Deep Learning Libraries and Architecture Choices

From DSC:  A library of Deep Learning methods.  This is quite technical, but a great idea.  Learning advances by not only learning ourselves but by sharing learning.  Unless we have a shared language and architecture that is very hard to do.  We used neural nets for machine learning in the enterprise, but stayed with single, simple architectures that replicated statistical methods we knew well.   Is the choice of DL method also possible to automate with cognitive methods?  Will consider that as I investigate the library.  Anyone researching this?