Incremental learning of temporal sequences using state memory and a resource allocating network
Yuichi Konishi, Robert H. Fujii · 2005
A network, which can learn temporal patterns incrementally, is proposed. The network uses: a) Gaussian functions to represent various output values and b) chunking to group similar patterns so that a compact network can be realized. The learning capacity limits of the network are evaluated for the worst case sequences. Learning of sequences can be performed using a simple one step algorithm.