Estratégia CDM: explorando consenso Fuzzy no gerenciamento de memórias híbridas

Abstract

This Thesis considers the consensual analysis in decision-making processes based on fuzzy logic (LF) and interval-valued fuzzy logic (IVFL), with the proposition of a strategy called CDM. In its foundation, the CDM, exploring the axiomatic properties of the fuzzy consensus measures (CM) formalizes two models. The first, applied to the fuzzy values lattice, enables the analysis of fuzzy information for the terms of a linguistic variable (VL), obtained by the composition of restricted equivalence relations, aggregation functions and fuzzy implications, including CM-constructors for convex sums, conjugation operators and dualities. A second model applied to fuzzy sets, building a new consensual analysis modeling the cohesion of several terms related to the same VL, and the coherence between fuzzy sets referring to the lowest and highest projections, characterizing the uncertainty spot interval extension. Such models decide considering relevance criteria or qualitative assessments, via selection of alternatives in the modeling of membership functions and provide support for two algorithmic strategies: FS-FCM, applied to fuzzy values; and, IVFS-FCM, with its application covering fuzzy sets. This Thesis aims to contribute to the management of hybrid memories proposed by the Intf-HybridMem approach, which explores the identification of access patterns to the two memories that make up the hybrid system, formed by volatile and non-volatile memories. The management employed in the Intf-HybridMem approach contemplates the use of IVFL. In this context, more specifically, the algorithmic strategies proposed in this Thesis help Intf-HybridMem decision-making on two fronts: (i) the FS-FCM strategy explores measures of consensus in values fuzzy from membership functions; and (ii) the IVFS-FCM strategy, which considers the inaccuracy inherent in the measures of input variables, read/write frequency and access recency to explore the fuzzy set consensus measures, in addition to including the recommendation of migration as output. The evaluations carried out point to the viability of employing the strategies proposed by this Thesis, in particular in the management of hybrid memories and, in general, in the continuity of the group’s researches that contemplate fuzzy modeling in decision-making processes.