Abstract
This article introduces an exploration of general Type-2 (GT2) fuzzy logic systems (FLSs) via Zadeh's (Z) GT2 fuzzy set (FS) definition, with a strong emphasis on advancing uncertainty quantification (UQ). At the heart of our contribution is the introduction of Z-GT2-FLS, formed through the integration of Z-GT2-FS with the α-plane representation. We show that the design flexibility of GT2-FLS is increased as it takes away the dependency of the secondary membership function definition from the primary membership function. For learning, we provide a solution to the curse of dimensionality problem alongside a method to seamlessly integrate deep learning (DL) optimizers. This article further presents a dual-focused Z-GT2-FLS within a DL framework, intending to learn Z-GT2-FLSs that are capable of achieving high-quality prediction intervals alongside high precision. In this context, we assign distinct roles for αk-plane-associated interval type-2 FLSs through a composite loss function. In addition, we extend the application of Z-GT2-FLS to predictive distribution estimation, proposing a DL framework to learn the inverse cumulative distribution function by predicting entire quantile levels. We first reformulate the output of Z-GT2-FLS to represent a quantile level function, thereby offering flexibility in generating desired quantiles through α-planes. For learning, we propose a simultaneous quantile learning method alongside an adaptation mechanism to enhance learning performance. Through comparative analyses, we show that the Z-GT2-FLS excels in UQ compared to its fuzzy and DL counterparts. The contributions of this study underscore the versatility and superior performance of Z-GT2-FLS, positioning it as a valuable tool for UQ.
Original language | English |
---|---|
Pages (from-to) | 314-324 |
Number of pages | 11 |
Journal | IEEE Transactions on Fuzzy Systems |
Volume | 33 |
Issue number | 1 |
DOIs | |
Publication status | Published - 2025 |
Bibliographical note
Publisher Copyright:© 1993-2012 IEEE.
Keywords
- Accuracy
- deep learning (DL)
- design flexibility
- general type-2 (GT2) fuzzy sets (FSs)
- prediction interval (PI)
- uncertainty quantification (UQ)