Most conventional associative memory systems beal with input patterns that are known in advance. On the contrary, the proposed system memorizes input patterns and recalls memorized patterns sequentially. In the system, a new measure, "reliability of the memorized pattern," is introduced, Using the measure, a problem of sequential learning is solved. Memorized patterns with high reliability are treated as important ones and are not removed from the memory of the system. They are recalled with high priority. Memorized patterns with low reliability are replaced by new important input patterns. In this way, sequential learning and recall in the on-line mode are possible by using the measure of reliability. The effectiveness of the proposed syystem is shown in a simulation.
展开▼