Key points are not available for this paper at this time.
How can neural networks learn the rich internal representations required for difficult tasks such as recognizing objects or understanding language?
Bengio et al. (Mon,) studied this question.