Optimised network for sparsely coded patterns
C. J. Pérez Vicente, D J Amit · Journal of Physics A Mathematical and General · 1989
The performance of attractor neural networks storing sparsely coded patterns has been shown to be greatly improved on shifting from the -1, +1 representation of neural states to the 0, 1 representation. Here the authors show that when this shift is considered as a special case of the transformation of the dynamical variables which depends on a continuous parameter, the value of the parameter can be chosen to improve the performance of the network even further for every value of the bias in the patterns.