k501

""" Deep Auto-Encoder implementation
	
	An auto-encoder works as follows:

	Data of dimension k is reduced to a lower dimension j using a matrix multiplication:
	softmax(W*x + b)  = x'
	
	where W is matrix from R^k --> R^j

	A reconstruction matrix W' maps back from R^j --> R^k