Tuning similarity-based fuzzy logic programs

We have recently designed a symbolic extension of FASILL (acronym of “Fuzzy Aggregators and Similarity Into a Logic Language”), where some truth degrees, similarity annotations and fuzzy connectives can be left unknown, so that the user can easily see the impact of their possible values at execution time. By extending our previous results in the development of tuning techniques not dealing yet with similarity relations, in this work we automatically tune FASILL programs by appropriately substituting the symbolic constants appearing on their rules and similarity relations with the concrete values that best satisfy the user's preferences. Firstly, we have formally proved two theoretical results with different levels of generality/practicability for tuning programs in a safe and effective way. Regarding efficiency, we have drastically reduced the exponential complexity of the tuning algorithms by splitting the initial set of symbolic constants in disjoint sets and using thresholding techniques. These effects have been evidenced by several experiments and benchmarks developed with the online tool we provide to verify in practice the high performance of the improved system. • Improved closure algorithm for synthesizing symbolic similarity relations attached to sFASILL programs. • Safe conditions on symbolic substitutions to transform sFASILL programs into FASILL ones. • Formal proofs of two theorems that relate the computational behaviour of FASILL programs derived from sFASILL ones before or after being instantiated with safe symbolic substitutions. • Efficient and safe tuning algorithms for instantiating sFASILL programs according users preferences. • Implementation issues, online tool and experimental evaluation with benchmarks.