Efficient Deep Approximation of GMMs

There is no theory yet as to whydeep neural networks (DNNs from here on) are as effectiveand efficient in practice as they evidently are.