A structure trainable neural network with embedded gate units: multidimensional input/output and its learning algorithm

In this paper, a synthesis and learning method for the neural network with embedded gate units and a multidimensional input is proposed. When the input is multidimensional, the gate functions are controlled in a multidimensional space. In this case, a hypersurface, on which the gate function is formed should be optimized. Furthermore, the switching points should be considered on the unit input. An initialization and control methods for gate functions, which optimize the hypersurface, the switching point and the inclination, are proposed. The stabilization methods proposed previously, are further modified to be applied to the multidimensional environment. The gate functions can be trained together with the connection weights. Discontinuous function approximation is demonstrated to confirm the usefulness of the proposed method