Dynamical features simulated by recurrent neural networks

The evolution of two-dimensional neural network models with rank one connec ting matrices and saturated linear transfer functions is dynamically equiva lent to that of piecewise linear maps on an interval. It is shown that thei r iterative behavior ranges from being highly predictable, where almost eve ry orbit accumulates to an attracting fixed point, to the existence of chao tic regions with cycles of arbitrarily large period. (C) 1999 Elsevier Scie nce Ltd. All rights reserved.