State changing of neural networks based on image correlation
Eugene I. Shubnikov · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 1996
The question of state changing for phase space and correlation space of neural networks based on image correlation is considered. The phase space is used for investigation of network behavior and stable. The network behavior is considered as mutual effect of excitation and inhibition of hypothesizes. The model of network behavior and phase portrait of two competitional hypothesis is presented. If a relative excitation and inhibition are linear functions the system is stable and solution is single. The correlation space is connected a network characteristics with pattern images, i.e. with learning process. The energy relief structure of correlation space for image simulation as random field is considered. The top values of absolute and relative depth of energy relief in true attractor point is calculated.