A new method for the construction of nonlinear adaptive filters called adaptive support vector regression is introduced for signal detection in noisy environments. Modification of support vector regression for on-line learning is motivated by the chunking approach and is based on repeated re-training of the filter parameters without loss of former estimates. Performance of the proposed method was found superior to the method using a Resource-Allocating RBF networks with Givens QR decomposition and pruning.
Go back