T2K2: A Type II KASER

Abstract

The transformational methodology described in this paper induces new knowledge, which may be open under any deductive process. The method of transposition is used to maintain a maximum size for the application as well as meta-rule bases. The "move to head" method is used by both the application and metarule bases for hypotheses formation. Whenever an application rule is fired, it is transposed on the transposition list and also moved to the head on the other list. If any meta-rule on a solution path individually leads to a contradiction on the application rule base, then the offending meta-rule is expunged. Then, when the system is idle enter dream mode, whereby rule i ⇒ rule j is generated by the 3-2-1 skewed twister as a candidate most-specific meta-rule. Candidate most-specific meta-rules are "cored" to create one generalization per candidate. These candidate meta-rules are tested for application to each rule in the application domain rule base. In order to be saved in the meta base, they may not map any existing rule in the application domain rule base to one having the same antecedent as another in this base, but a different consequent (as found by hashing). In addition, all candidate meta-rules must map at least one rule in the application base to another distinct one there, or be symmetrically induced from meta-rules that so map.