A coding procedure for real-valued patterns using bidirectional associative memory

Abstract EN

A bidirectional associative memory (BAM), reported by Kosko, can be used ،0 store and recall binary /bipolar pattern pairs.

An iterative learning algorithm called PRLAB has been used for guaranteed recall of all training pairs up to the maximum storage capacity of a BAM.

In this paper we propose a new coding strategy for storage and recall of real- valued patterns using BAM.

The proposed algorithm is based on the PRLAB iterative learning algorithm, which enables the BAM to be applied in industrial environments where analog patterns are commonly encountered.

An algorithm has been developed for converting real-valued pattern pairs into binary/bipolar patterns, thus enabling it to be applied to a discrete BAM.

The PRLAB learning algorithm is then applied to store and recall the real- valued pattern pairs successfully.

£xamples are given demonstrating the applicability of the new encoding mechanism for real-valued patterns with sufficient resolution to be used in practical application.

-