Analysis and synthesis of a continuous-time hysteresis neural network
Kenya Jin’no, Toshimichi Saito · 2003
A piecewise-linear hysteresis associative memory is discussed. Conditions on parameters for guaranteed storage of all desired memories, global convergence of an energy function, and control of stable spurious output are given. The networks are then synthesized using a mixed autocorrelation/pseudoinverse matrix. In some range of their mixed rate, it is numerically confirmed that the performance is improved vastly. A remarkable hysteresis effect in the global convergence properties of each desired memory is found.>