Bidirectional Associative Memories for Adaptive Rule Based Systems

In this paper we present an algorithm for rule based systems that uses Bidirectional Associative Memories (BAMs) to memorize the inputs and their corresponding outputs, and outputs and their corresponding inputs of the system. Adaptive rule based system are becoming very important because rule based systems are now used in many applications. One drawback of current adaptive rule based systems is that these systems care only about forward chaining mechanism which diminish their performance. Because most of the applications that use rule based systems follow the forward chaining, adaption attempts which is concerned with predicting the inputs given the output is almost null, but this does not eliminate the importance of backward chaining since it is used in too many applications. To tackle this problem we propose a new algorithm that utilizes the good theoretical ground of BAMs to memorize and adapt rules in rule based systems. The main difference between the proposed algorithm and other algorithms is that the proposed algorithm is simple to code the rules and adapt them. In addition, because BAMs supports bi-directions, the proposed algorithm is the only algorithm with adaptation that is able to expect what conditions should be true if we provide the system with outputs. The proposed algorithm considers "and" and "or" rules in the used rules, whereas in all the previous systems, only "and" relation is considered. The proposed solution will be useful in adapting any rule based system whether it uses forward chaining or backward chaining, which is considered to be a significant contribution. The proposed algorithm has six parts; cod table creation, input and output vectors construction, weight matrix calculation, procedure to test the system, procedure to run the proposed algorithm and finally rule extraction or what we call rule code decoding. To illustrate parts of the proposed algorithm, a simple example is provided.