Behavior stabilization of complex-valued recurrent neural networks using relative-minimization learning

Relative-minimization learning using additional random teacher signals is proposed for recurrent-behavior stabilization. Although the recurrent neural networks can deal with time-sequential data, they tend to show an unstable behavior (positive Lyapunov exponent). The proposed method superimposes a type of basin upon a dynamics-determining hypersurface in an information vector field. This process is equivalent to the relative minimization of the error function in the input-signal partial space. Experiments demonstrate that the relative-minimization learning suppresses positive values of Lyapunov exponents down to zero or negative, resulting in a successful behavior stabilization