A combined network architecture using ART2 and back propagation for adaptive estimation of dynamical processes

Quantitative object motion prediction by an ART2 and Madaline combined neural network

“…The simulation results of the combined network are compared to a feed forward network using a back propagation learning [6] 2 Table 1 lists the EMse values for the test cases, where cases S1 to $4 use artificially generated data and case AA uses actual object motion profile. It shows that the combined network improved the prediction precision over the back propagation network in all testing cases.…”

“…Eisner and the others [4,5,6] have studied the use of artificial neural networks to predict time series, a problem very close to the motion prediction. A back propagation network was used by Gent [5] for the prediction of both deterministic and stochastic time series.…”

“…Sorheim utilized an ART2 and back propagation combined network for the estimation of dynamic processes [6]. However, the slow speed of the back-propagation learning makes the networks not suitable to the motion prediction which requires a timely adaptation to abrupt and irregular changes of the object motion.…”