In research published in the International Journal of Machine Learning and Cybernetics, investigators Xiaoyu Peng and Xiaodong Pan introduced a novel architectural framework for interval type-2 fuzzy systems that utilizes parametric n-ary 1-overlap functions as conjunction operators. Fuzzy logic architectures are established engineering tools designed to model pervasive uncertainty across complex environments, ranging from dam displacement forecasting and electrical circuit control to permanent magnetic drives. Although interval type-2 fuzzy sets offer greater capacity to handle severe noise and uncertainty than conventional type-1 systems while maintaining greater operational simplicity than general type-2 configurations, traditional inference engines remain constrained by the mathematical limitations of standard conjunction operators. By replacing standard triangular norms with flexible parametric n-ary 1-overlap functions, the authors demonstrated that interval type-2 fuzzy systems can achieve concise operational structures, preserve essential continuity, and deliver verified universal approximation capabilities. Experimental validation confirmed that integrating these parametric operators significantly reduces approximation errors across multidimensional nonlinear function benchmarks.
To satisfy the stringent precision demands of modern engineering systems, conventional fuzzy inference mechanisms frequently couple rule bases with automated parameter optimization algorithms. In standard fuzzy implementations, the primary approach involves adjusting the geometric parameters that define the input membership functions. However, systematically altering antecedent membership functions during training often causes an irreversible loss of the domain-specific expert knowledge initially embedded within those linguistic terms. To prevent this degradation, researchers previously proposed adjusting the conjunction operators that connect rule antecedents, exploring parametric triangular norms or t-norms to adjust system behavior. Nevertheless, standard t-norms are bound by the mathematical requirement of associativity. This algebraic restriction frequently forces parametric t-norms into highly complex analytical formulations that substantially increase computational overhead during runtime inference. Importantly, theoretical studies have established that associativity is not a mandatory property for successful fuzzy system design, creating an opportunity to construct non-associative operators that maintain computational efficiency.
To overcome the structural bottlenecks of associative operations, the authors adopted n-ary 1-overlap functions, a class of non-associative aggregation operators initially proposed to solve multi-dimensional classification and image processing threshold problems. Unlike classical t-norms, an n-ary 1-overlap function is defined as a continuous, symmetric, and monotonically increasing mapping from the unit hypercube to the unit interval that evaluates to zero if and only if at least one input coordinate is zero, while reaching unity when the coordinate product equals one. Beyond structural conciseness, n-ary 1-overlap functions possess a wider operational output range than t-norms. In traditional t-norm formulations, the aggregated result cannot exceed the minimum antecedent value, which restricts the model to capturing solely weakening interactions between rule antecedents. In contrast, parameterized 1-overlap functions can generate values greater than the minimum of their inputs, properly modeling synergistic antecedent reinforcement as illustrated by the classic watermelon problem, where combining moderate evidence of a red interior and a green exterior yields a strengthened overall classification score.
A vital mathematical characteristic of n-ary 1-overlap functions explored in the study is their linear closure under convex combinations. According to Theorem 1, any convex combination formed from multiple n-ary 1-overlap functions using non-negative weights that sum to unity yields another legitimate n-ary 1-overlap function. Classical t-norms do not possess this closure property, which severely restricts their ability to blend diverse aggregation behaviors smoothly. Capitalizing on this advantage, the authors introduced a composite parametric operator designated as O_MP, which forms a weighted convex combination of a parameterized minimum operator O_M and a parameterized product operator O_P. Governed by a scalar weighting parameter alongside power parameters for the individual components, this composite operator provides a simple, flexible mechanism to tune the balance between distinct operational behaviors. During optimization, tuning this weighting parameter allows the system to seamlessly transition between different conjunction characteristics, identifying optimal interaction dynamics for specific modeling problems without requiring complex, piecewise mathematical formulations.
The construction of the proposed system begins with the generation of multi-dimensional interval type-2 fuzzy sets derived from underlying vague partitions. The researchers formulated both cosine and triangular vague partitions over real input intervals, dividing the continuous universe of discourse into overlapping fuzzy regions that satisfy continuity, unimodality, and partition-of-unity conditions. To capture antecedent uncertainty, the authors introduced a perturbation mechanism based on polynomial functions, which expands each one-dimensional type-1 fuzzy set into an interval type-2 fuzzy set bounded by explicit upper and lower membership functions. These individual one-dimensional fuzzy representations were subsequently combined into multivariate antecedent sets on multi-dimensional input domains through the application of n-ary 1-overlap functions. By aggregating the lower membership boundaries with a lower overlap operator and the upper boundaries with an upper overlap operator, the framework establishes a consistent multi-dimensional footprint of uncertainty that preserves the underlying structural topology while providing tunable parameters for subsequent optimization.
Using the constructed multi-dimensional fuzzy sets, the authors formulated a dual-input single-output interval type-2 fuzzy system model and established its fundamental theoretical foundations. In the defuzzification stage, standard interval type-2 systems traditionally apply Karnik-Mendel algorithms to compute left and right boundary centroids, taking their unweighted arithmetic mean as the crisp output. In contrast, the authors introduced an explicit preference degree parameter, sigma, yielding a generalized convex combination of the boundary outputs. Adjusting this parameter allows the system to transition smoothly between conservative and aggressive inference behaviors. Crucially, the authors mathematically proved that this architecture satisfies the interpolation property, meaning that at the center points of the antecedent partitions, the system output precisely matches the corresponding rule consequence. This property holds because n-ary 1-overlap functions vanish exclusively when coordinate products are zero. Building upon this interpolation result, the researchers proved that the architecture possesses first-order approximation and universal approximation capabilities for continuous functions.
To systematically calibrate system parameters, the authors structured three distinct optimization models termed IT2FS-PO1, IT2FS-PO2, and IT2FS-PO3. The first model, IT2FS-PO1, incorporates adaptive nodes exclusively within the inference engine to determine the parameters of the n-ary 1-overlap functions. The second model, IT2FS-PO2, extends adaptability to the defuzzification stage by simultaneously optimizing the conjunction parameters and the preference degree parameter sigma. The third model, IT2FS-PO3, employs a two-stage sequential optimization strategy that first optimizes the inference and defuzzification parameters, and subsequently tunes the antecedent membership function parameters while holding the previously calibrated values fixed. Furthermore, two distinct parameter allocation schemes were examined: Model1, which shares conjunction parameters across all rules, and Model2, which assigns independent, rule-specific parameters to every individual fuzzy rule. Parameter search was executed using the Quantum-behaved Particle Swarm Optimization algorithm, an iterative global search technique that eliminates partial derivative calculations and demonstrates rapid convergence without requiring velocity vector adjustments.
The empirical performance of the proposed architecture was initially evaluated through numerical approximation experiments on the two-dimensional Schaffer benchmark function f_1 across 121 uniformly distributed sample points. Performance was measured using root mean square error and an improvement percentage index relative to a classical interval type-2 baseline utilizing a standard product operator. The simulation results revealed that Model1 configurations provided modest error reductions due to having very few adjustable parameters. Conversely, transitioning to Model2, where parameters vary across individual rules, yielded dramatic performance improvements across all tested configurations. Furthermore, the composite operator O_MP consistently outperformed both individual minimum and product operators, illustrating the practical benefit of convex linear combinations. Incorporating the tunable preference parameter sigma generated substantial performance gains, particularly in configurations with fewer overall parameters. When antecedent membership functions were subsequently adjusted in the IT2FS-PO3 model, improvement percentages exceeded 60 percent relative to the classical baseline.
To verify generalization capability, the authors extended experimental testing to five additional standard benchmark functions: Schaffer Function N.4, Schaffer Function N.2, the Gramacy and Lee function, the Matyas function, and the Holder Table function. The proposed IT2FS-PO3-Model2 architecture, implemented with cosine membership functions on a five-by-five partition and the composite overlap operator, was benchmarked against alternative fuzzy techniques as well as popular machine learning algorithms. Comparisons against the established A2-C1 interval type-2 model revealed the marked superiority of the proposed framework, largely because the Gaussian membership functions and product operators in A2-C1 fail to satisfy the interpolation property, leading to degraded performance when using compact rule bases. When evaluated alongside standard machine learning algorithms—including Support Vector Regression, Random Forest, Gradient Boosting, Bagging, and K-Nearest Neighbors—the proposed fuzzy system achieved competitive root mean square error and mean absolute error scores, demonstrating robust stability and strong generalization across diverse analytical surfaces.
In their concluding assessment, the researchers emphasized that the developed framework provides a rigorous theoretical and methodological foundation for designing high-performance interval type-2 fuzzy systems. For applications where engineering domain knowledge must be strictly preserved without modifying expert-defined membership functions, the IT2FS-PO2-Model2 configuration serves as an optimal solution by confining optimization to rule-specific aggregation operators and the defuzzification preference parameter. Nevertheless, the authors highlighted important limitations that contextualize their present findings. The current experimental verification relied exclusively on synthetic mathematical benchmark functions, which do not fully replicate the complex, irregular noise structures encountered in real-world physical processes. Consequently, applying these parametric overlap models to challenging empirical datasets under real-world operating constraints represents the central objective of ongoing research. Additionally, future investigations will examine whether integrating parametric overlap indices into interval type-2 architectures can further enhance model flexibility and systemic precision across broader automated control environments.
Subject of Research: Interval type-2 fuzzy logic systems employing parametric n-ary 1-overlap aggregation functions
Article Title: Interval type-2 fuzzy systems based on parametric n-ary 1-overlap functions
Article References: Peng, X., & Pan, X. (2026). Interval type-2 fuzzy systems based on parametric n-ary 1-overlap functions. International Journal of Machine Learning and Cybernetics, 17(12), Article 546. https://doi.org/10.1007/s13042-026-03376-y
Image Credits: AI Generated
DOI: 10.1007/s13042-026-03376-y
Keywords: interval type-2 fuzzy systems, fuzzy logic, n-ary 1-overlap functions, aggregation functions, universal approximation, quantum-behaved particle swarm optimization, parametric operators, information fusion, vague partitions, defuzzification, Karnik-Mendel algorithms, function approximation, computational intelligence, Schaffer benchmark function, interpolation property, root mean square error, fuzzy neuro-systems
Cite Scienmag News
Courtney Benton. (November 20, 2026). Parametric Overlap Functions Enhance Accuracy in Interval Type-2 Fuzzy Logic Systems. Scienmag. https://scienmag.com/parametric-overlap-functions-enhance-accuracy-in-interval-type-2-fuzzy-logic-systems/
Courtney Benton. "Parametric Overlap Functions Enhance Accuracy in Interval Type-2 Fuzzy Logic Systems." Scienmag, 20 November 2026, https://scienmag.com/parametric-overlap-functions-enhance-accuracy-in-interval-type-2-fuzzy-logic-systems/. Accessed 20 November 2026.
Courtney Benton. "Parametric Overlap Functions Enhance Accuracy in Interval Type-2 Fuzzy Logic Systems." Scienmag. November 20, 2026. https://scienmag.com/parametric-overlap-functions-enhance-accuracy-in-interval-type-2-fuzzy-logic-systems/








