Representation of information in a neural network using psychophysical functions and the maximum entropy formalism

Summary form only given, as follows. The authors explored the possibility that the correct encoding of the information at the input-layer level in a neural network (not at the hidden-layer level, as usually assumed) is a requisite for its correct representation. They proposed a mechanism that calculates the psychophysical function of the input data to obtain the canonical coordinates with which the network will operate and then filters the resulting values using a maximum entropy algorithm to eliminate the spurious information that inevitability arises using psychophysical functions. They considered the possible implications of the proposed model, especially the last part, which could be viewed as a primitive model of consciousness.