In this section the ability of the algorithm to learn sparse codes is verified in simulations. Section 4.3.1 shows what are the outputs of the network, when the weight vectors are fixed. The outputs will look very much like those presented in figure 3.1. Sections 4.3.2 and 4.3.3 show that the algorithm is able to learn meaningful sparse representations from artificial and natural data.

Thu May 9 14:06:29 DST 1996