It is shown that training feed-forward nets can be viewed as a system identification problem for a nonlinear dynamic system. For linear dynamic systems, the Kalman filter is known to produce an optimal estimator. Extended versions of the Kalman algorithm can be used to train feed-forward networks. The performance of the Kalman algorithm is examined using artificially constructed examples with two inputs, and it is found that the algorithm typically converges in a few iterations. Backpropagation is used on the same examples, and the Kalman algorithm invariably converges in fewer iterations. For the XOR problem, backpropagation fails to converge on any of the cases considered, whereas the Kalman algorithm is able to find solutions with the same network configurations.>

VLSI architecture design and implementation of a chip pair for the motion compensation full search block matching algorithm are described. This pair of ASICs (application-specific integrated circuits) is motivated by the intensive computational demands for performing motion compensation in real time. They have been developed to calculate fractional motion vectors with quarter-pel precision. The VLSI architecture is based on some special data-flow designs that allow sequential inputs but perform parallel processing with 100% efficiency for integer motion vector estimation and nearly 100% for fractional motion vector estimation. The chip-pair design has been laid out and simulated using a silicon compiler tool, and the chip statistics are summarized. Testing circuitry is included to increase the observability of the chips.>