RC. The feedforward NN is a classifier network structure, in which each unit in the hidden and output layers may be configured as shown in Fig. 8. Typically, the learning scheme used during training to adjust the connection weights is some form of back propagation of squared errors between the observed and target outputs. Several authors have made connections between classifier networks and statistical pattern recognition.
Radial basis function. 2 The bipolar SF The hyperbolic tangent is just a specific bipolar SF (BSF) with a continuous range from -1 to +1 as shown in Fig. 4. 1b) i=1 ~ [WI W2, ... , Wn]T and i repre~ [il i 2 , ... , in]T. 50 '0 c. 25 -10 10 0 x Fig. 4 Shape of the bipolar continuous function. 1) that the SF fO, as a processmg neuron, performs the operation of summation of its weight inputs. 3) where a > 0 is the BSF gain (parameter) for determining the steepness of the continuous BSF, as shown in Fig.
Intelligent Control Based on Flexible Neural Networks by Mohammad Teshnehlab, Keigo Watanabe (auth.)
by George
4.5