Approximate case-based reasoning on neural networks

Zuliang Shen, Ho Chung Lui, Liya Ding
1994 International Journal of Approximate Reasoning  
In this paper, based on the deeper analysis of the features of fuzzy logic and approximate reasoning, the concept of approximate case-based reasoning (ACBR) is introduced. According to the inference mechanism of ACBR, an implementation on neural networks is proposed. Mapping the implication relation between the premise(s) and the consequence of a fuzzy rule to the weight of a corresponding neural network unit, an approximate case-based reasoning on neural networks can be realized. The
more » ... izing and self-learning procedure can be executed by modifying the weight. Zuliang Shen et al. true, we prefer to consider them as examples or case-rules rather than rules. The feature is called the rule-case duality of fuzzy rules in approximate reasoning. In the sense, an approximate reasoning done by using those fuzzy rules is the approximate case-based reasoning. Another assumption in binary logic is that any proposition can take only the truth values 1 or 0, where 1 means true and 0 means false for positive logic (for negative logic, vice versa). Usually, we write a true proposition as itself, such as P, and a false proposition as the negation of a true one, such as ~ P. The logical reasoning in binary logic can always be symbolically processed because all the propositions take the unique truth value 1. Different from binary logic, there are infinitely many possible truth values in fuzzy logic; any truth value between 0 and 1 (fuzzy-valued logic) or any linguistic truth value in [0, 1] 2 (fuzzy linguistic-valued logic) [4, 5] . So fuzzy logic introduces numbers into the pure symbolic system by using fuzzy truth and linguistic truth values. However, any fuzzy proposition in either fuzzy-valued or fuzzy linguistic-valued logic can be represented by a true proposition (that is, in fuzzy-valued logic it can be considered as truth value 1) with some kind of confidence degree. Hence, the true proposition can still be dealt with as a symbolic term and its confidence degree is processed as a numeric value [6] . This feature is called the symbolic-numeric duality of fuzzy logic.
doi:10.1016/0888-613x(94)90010-8 fatcat:4mjkamilzncptmekgg675hyt5u