Artificial intelligence

Abstract: Practical applications of q-state Potts models are complicated, as they require very large RAM (32N^2q^2 bits, where N is the number of neurons and q is the number of the states of a neuron). In this work we examine a modified Potts model with binarized synaptic coefficients. The procedure of binarization allows one to make the required RAM 32 times smaller (N^2q^2 bits), and the algorithm more than q times faster. One would expect that the binarization worsens the recognizing properties. However, our analysis shows an unexpected result: the binarization procedure leads to the increase of the storage capacity by a factor of 2. The obtained results are in a good agreement with the results of computer simulations.

Keywords:

References: