Temporal pattern identification in a self-organizing neural network with an application to data compression

Summary form only given. A neural network system is presented that seeks to find the optimal structure embedded in a temporal sequence. This structure is manifested by the existence of a finite number of subsequences that comprise the temporal sequence. The measure for optimality is the information rate and it is precisely defined and utilized in the network. The network consists of three levels of processing units: the first level grows a tree based on the inputs, the second selects groups of nodes in the tree to form sets of subsequences that completely cover the input, and the last calculates the information rate and indicates which set is the most optimal. Experimental simulations reveal various attributes of this system. An application has been made to data compression