PROBABILISTIC PROPERTIES OF THE INITIAL VALUES OF WEIGHTING FACTORS IN SYNCHRONIZED ARTIFICIAL NEURAL
В. Ф. Голиков, Н. В. Брич · DOAJ (DOAJ: Directory of Open Access Journals) · 2015
One of the most efficient ways for identical binary se quences generation is using methods of neural cryptography. The initial weight vestors values influence on speed of synchronization is analized. Equal probability of initial weight vestors motion directions is great advantage. On this base authors suppose new line of research conserned with improvement of network architecture and correction algorithm.