Seminar of Theoretical Computer Science - Dušan Guller (13.4.2018)
Friday 13.4.2018 at 11:00, Lecture room M/213
By: Rastislav Kr8lovič
On Multi-step Fuzzy Inference in Goedel Logic
Recent years witness a prospective application potential of the fuzzy logic and inference approach to emerging technologies in the area of artificial, computational intelligence, and soft computing. Fuzzy rules and one-step fuzzy inference is widely exploited in fuzzy controllers since the seventies. From a viewpoint of artificial intelligence, such a kind of inference has a reactive behavior. In contrast to control purposes, one-step fuzzy inference is not sufficient for so-called fuzzy reasoning, where some kind of abstract inference is needed to reach a reasonable conclusion, which stipulates multiple inference steps.
In our talk, we will discuss the logical and computational foundations of multi-step fuzzy inference using the Mamdani-Assilian type of fuzzy rules by implementing such inference in Goedel logic with truth constants. We apply the results achieved in the development of a hyperresolution calculus for this logic. We pose three fundamental problems: reachability, stability, the existence of a k-cycle in multi-step fuzzy inference and reduce them to certain deduction and unsatisfiability problems. The corresponding unsatisfiability problems may be solved using hyperresolution.