Deep learning is the latest attempt to add intelligence to computational processes and has resurrected that old chestnut AI. The term ‘deep’ refers to the number of layers within the learning process and isn’t some allusion to the profundity of the methods. Neural networks usually implement two or three layers, and going any deeper than this has traditionally resulted in very poor models. A paper by Geoffrey Hinton in 2006 dealing with Deep Belief Networks changed the playing field and made many layer learning structures feasible.
In the ongoing quest to emulate the human brain (which most neurologists will admit is hardly understood at all) it has been shown that the brain has a deep architecture – and particularly well studied is the visual cortex which has multiple levels of processing.
The algorithms that implement deep reasoning are still quite new and are mostly implemented in Python. Most tools are libraries of C++ code, Python modules or Matlab code – so it’s not particularly plug and play. Theano is one of the more popular tools and is a symbolic expression compiler in Python. Much more can be learned at deeplearning.net.