A new connectionist model for associative retrieval of harmonic tunes

초록

Many of the existing connectionist models based on neural associative memory have two separate stages for learning and retrieval, and are not structurally suitable for temporal pattern storage and classification in that the user must properly specify the duration of unit sector of continuously incoming patterns. In this paper, a new connectionist model of on-line associative memory for temporal pattern store and retrieval is suggested. The input signal that the suggested model deals with is a set of multi-channel discrete sequences of binary patterns which can be an interpretation of musical tune played on a keyboard. Simulation shows the suggested model can associatively retrieves a few stored patterns and adaptively tunes its stored pattern upon the change of input patterns.