FCTA 2026 Abstracts

Full Papers Short Papers
Full Papers
Paper Nr: 26
Title:

On Parametric Extensions of Fuzzy Connectives: The Alpha, Beta, Gamma Operators
Authors:

József Dombi and Noor Bahjat Tayfor
Abstract: In fuzzy theory, many operators have been proposed and extended so far. From a practical perspective, the operators are continuously differentiable and analytic in practical applications. Due to flexibility reasons, we need operators with parameters. In our previous study, we established that the generalised forms of the α and γ operators include the most popular ones. This operator has only two parameters. The conjunctive and disjunctive operators differ only in the sign of the α parameter. Furthermore, it was shown that the product, Einstein, Hamacher, min-max, drastic and Dombi operators can be derived as special cases. Moreover, the min-max and Dombi operators satisfy the De Morgan identity for all rational strong negations. Here, we introduce an additional β parameter and derive the so-called α, β, γ operator, which is a strictly monotone operator. We will also discuss the two special cases in Dombi and product operators. Additionally, we will derive several additional operators as special cases by adjusting the values of the α and γ parameters. Finally, we will use the Archimedean property to obtain a specific approximation that recovers the originally proposed parameter values of α, β, and γ.
Paper Nr: 29
Title:

Efficient Encodings of the Uncertain MIN Connector for Possibilistic Weighted Model Counting
Authors:

Guillaume Petiot
Abstract: Uncertain logical gates are powerful tools for knowledge representation. They enable the automatic computation of conditional possibility tables in possibilistic networks, using a limited set of parameters and a function that defines their behavior. For binary variables, classical logical operators such as OR, AND, and NOT can be used as connector functions. However, multivalued variables require a broader class of connectors. In this study, we focus on the uncertain AND connector and its generalization, the uncertain MIN connector. We investigate the encoding of these connectors for possibilistic weighted model counting (π-WMC). Such encodings can be directly integrated into the global encoding of a possibilistic network, thereby leveraging the advanced optimizations provided by solvers. We further introduce compact query-specific encodings (QSE), including a hybrid max-precomputation strategy tailored to query-specific inference. Finally, we experimentally compare several encodings and show that QSE-based approaches significantly improve scalability for large inference problems in possibilistic reasoning systems.
Paper Nr: 140
Title:

Harnessing the Extensibility of a Product Fuzzy SAT Solver
Authors:

Ivor Uhliarik
Abstract: Product fuzzy logic remains relatively underrepresented in the field of automated theorem proving despite substantial advances in satisfiability solving in other fuzzy logics. Existing approaches on the propositional level are based on stochastic techniques, translations to satisfiability modulo theories and mathematical optimization problems, or fuzzy extensions of classical SAT solving methods. We present improvements to the existing DPLL-motivated solver for propositional product fuzzy logic extended with the Monteiro--Baaz $\del$ connective and order connectives. The proposed enhancements consist of a simplification step and an additional reduction rule that decrease the search space and improve solving efficiency. Their effectiveness is evaluated using a previously established test bench and compared with earlier versions of the solver, where results exhibit a drop in runtime to $47.88\%$ in one of the experiments. We further demonstrate the extensibility of the solving approach by incorporating a new DPLL-like rule for a class of computationally hard problems. Although not intended as a general-purpose optimization, the heuristic substantially reduces the search space and solving time that beats state of the art, illustrating the benefits of a transparent, rule-based architecture for product fuzzy satisfiability solving.
Short Papers
Paper Nr: 84
Title:

Precise Formulations of Fuzzy Non-Linear Regression System
Authors:

Kiyoshi Nagata
Abstract: Regression analysis models in statistics are extremely use- ful for data prediction, decision-making, and many other applications, and are used in numerous fields. Even for relatively simple multivariate linear models, issues such as data standardization or the selection of vari- ables are being studied, and identifying the optimal nonlinear function is also a problem when applying them to nonlinear models. When the data is quantitative, research methods have been established to some extent through application software and empirical research. In particu- lar, regression analysis models using kernel methods and machine learn- ing techniques have been proposed and are being utilized for generating AI. In contrast, when the data is fuzzy, containing ambiguity, several researchers have proposed for regression analysis models, and some em- pirical research has been conducted in certain fields, but the situation does not seem to be mature. This research aims to reconsider regres- sion analysis models for fuzzy data using machine learning methods and give a price formulation which may enable us to construct an application program treating a data set of fuzzy numbers.
Paper Nr: 109
Title:

An AI-Driven Plugin-Based Web Platform for Two-Way Circular Intuitionistic Fuzzy Analysis of Variance
Authors:

Velichka Traneva, Petar R. Petrov and Stoyan Tranev
Abstract: Healthcare data analysis frequently involves incomplete and uncertain information, which may limit the ability of conventional statistical methods to explicitly represent uncertainty arising from professional assessments. Although classical analysis of variance (ANOVA) is widely used to investigate differences associated with categorical factors, it does not directly incorporate hesitation or disagreement among assessors. This paper presents an AI-driven pluginbased web platform implementing Two-Way Circular Intuitionistic Fuzzy Analysis of Variance (C-IFANOVA) for uncertainty-aware statistical decision support in healthcare. The platform integrates data preprocessing, professional assessment, Circular Intuitionistic Fuzzy transformation, variance analysis, visualization, and report generation within a unified analytical environment. Independent assessments provided by medical professionals are aggregated and represented as Circular Intuitionistic Fuzzy Triples (C-IFTs), whose uncertainty radii characterize the dispersion among the professional assessments. The proposed framework is evaluated using the publicly available Diabetes Health Indicators dataset to investigate whether BMI category and age group are associated with statistically significant variation in diabetes risk. The obtained results are consistent with those of classical Two-Way ANOVA, while additionally preserving information about membership, non-membership, hesitation, and expert disagreement. The conclusions remain stable across the considered aggregation strategies and uncertainty scenarios. The modular plugin-based architecture enables the integration of additional statistical and uncertainty-aware analytical methods and supports extension of the platform to other structured healthcare and engineering applications.
Paper Nr: 119
Title:

Adversarial IoT Sensor Attacks in Smart Cities: A Robustness Study of Fuzzy Regression Trees for Urban Traffic Prediction
Authors:

Mattia Daole, Giustino Claudio Miglionico, Pietro Ducange, Michela Fazzolari and Francesco Marcelloni
Abstract: Smart cities rely on Internet of Things-based sensing infrastructures to support mobility services and traffic management. In this context, traffic-speed forecasting is crucial, but its accuracy and explainability depend on sensor measurements whose integrity cannot always be guaranteed. Faulty or manipulated readings may corrupt model inputs at inference time, degrading predictions and explanations. This paper investigates inference-time sensor-integrity attacks on traffic-speed forecasting, focusing on fuzzy regression trees as interpretable predictors. The fuzzy regression tree is compared with post-hoc explainable black-box models under noise injection, spoofing, and drift-based attacks. Forecasting accuracy and explanation reliability are jointly evaluated. The results show that spoofing and persistent drift are the most damaging attack families, while the fuzzy regression tree provides a favorable trade-off between robustness and interpretability.
Paper Nr: 101
Title:

Fuzzy Linear Regression through the Extension Principle: A Comparison of Two Approaches
Authors:

Aurélien Pardo and Laurent Donzé
Abstract: Classical regression assumes precise data, which is unsuitable when relationships and quantities are inherently vague. Fuzzy linear regression addresses this by allowing both variables and parameters to be fuzzy numbers, restoring estimation reliability. In this article, we study two approaches to the fully fuzzy model grounded in Zadeh’s extension principle: the theoretically established method of Wu [11], which analytically derives the membership functions of the regression coefficients, and the resampling-based method of Rosset and Donzé [8], in which the fuzziness encoded in the data is itself resampled to estimate the coefficients. The aim is to validate this distribution-free, easyto-implement resampling method against the benchmark provided by Wu’s approach. To ensure a fair comparison, the following new contributions are proposed. First, a Bound Profile Enumeration (BPE) scheme that exhaustively explores all bound combinations in Wu’s approximation is developed and implemented. For Rosset and Donzé’s method, a quantile-based clipping rule is introduced, addressing the sensitivity of the pmin,maxq rule to extreme values. A principled criterion for selecting the number of resamples, derived from the Dvoretzky–Kiefer–Wolfowitz inequality and Monte Carlo simulation, is also proposed and applied. The two methods are compared on Wu’s original dataset and on the easySHARE dataset, using visual inspection and fuzzy-number metrics. The resampling approach yields coefficients consistent with Wu’s while producing wider supports that typically encompass them, reflecting a richer modelling of imprecision, which on easySHARE helps characterise the target variable, selfperceived health in the United States (sphus). These results establish the resampling approach as a credible, practical, and distribution-free alternative in the fuzzy regression toolbox.