Just a note: it's only my hobby! :)
There are various types of NN: MLP, RBM, DBN, DBM.
Everything is trained with gradient descent, there is also greedy layer-wised unsupervised pretraining algorithm :)
Each network has it's own manager, you just need to provide neural network heuristic and database.
There is also supervised and unsupervised benchmarking of networks, so it should be easy to compare networks.
I've created fast database for providing generic data-vectors concurrently for neural networks.