In this study, we elaborate on the idea of parametric group consensus measures. Here, we employ the additive generators of Dombi t-norms to construct fuzzy entropies, which are then utilized to generate group consensus measures. Since these additive generators are single-parameter functions, the resulting group consensus measures are also single-parameter mappings. We demonstrate that, given a set of inputs from decision-makers, the group consensus measure behaves as a bounded and non-decreasing function of its parameter. With this in mind, we address the question of which parameter value yields a group consensus measure that most accurately reflects the average perceived consensus level of the group. Next, we present a necessary and sufficient condition for the existence of an optimal value of the parameter in question, and provide algorithmic procedures to find it.
The Search and Rescue Problem (SARP) can be formulated in an environment subject to both objective uncertainty (randomness inherent to nature) and subjective uncertainty (lack of knowledge about the state of the world). In this paper, we present an interval arithmetic interpretation of the uncertainty problem. A crisis scenario is modeled as an assignment and optimization problem with interval-valued parameters and constraints. These intervals capture uncertainty over the problem data. A branch and bound algorithm is used to explore the solution space. Interval arithmetic is employed to compute bounds and obtain feasible assignments. From the resulting assignments, residual injury intervals are derived to assess the impact of uncertainty on each wounded person. Parallel computing techniques are also investigated to reduce execution times in the solution process.
Verifying timed systems is essential in safety-critical applications and poses significant challenges due to the huge number of possible behaviors. Model checking is a powerful tool of verification exploring the possible behaviors of systems, but it struggles with industrial applications, particularly when time is involved. Abstraction-based techniques are often used for the model checking of complex systems, as they can simplify the representation of the state space. Lazy abstraction is such a technique that incrementally adjusts (refines) abstractions only as needed, thus offering a promising solution by balancing precision and efficiency. This paper explores both sequential and parallel implementations of lazy abstraction strategies for timed systems. The sequential approach generalizes former implementations, while parallel implementations aim to leverage modern multi-core architectures for improved scalability and efficiency. Through empirical evaluation of benchmark systems, the research compares the various approaches, identifying key factors influencing parallelization effectiveness. The findings offer insights into lazy abstraction and opens new directions to further enhance the verification of complex timed systems.
Interval observers typically rely on performance criteria that specify the estimation accuracy. Inappropriately parameterized performance criteria can lead to linear matrix inequalities (LMIs) that do not yield feasible solutions. Retuning the output specification of the error dynamics allows for exploiting structural feasibility. The overall goal of this paper is to design a TNL interval observer for systems for which the LMIs do not yield a feasible solution with the standard specification of the performance criteria. As an example, the proposed tuning methods are applied to the state estimation of a lithium-ion battery cell. The LMIs introduced together with this observer structure initially do not have a feasible solution for the investigated battery model. To obtain feasible solutions, we propose extended design conditions for the TNL interval observer based on a virtual output equation for the error dynamics that is utilized to reduce the influence of uncertainties on the estimation results. Additionally, we present two techniques to enforce interpretable structures in the observer gains to enhance the estimation accuracy. The first technique enforces a desired ratio between the elements of selected observer gains, the second technique influences the estimation accuracy by specifying the eigenvalues of the scaled observer system matrix. To demonstrate the fundamental application of the tuning methods, the proposed techniques are initially applied to the state estimation of a mass-spring-damper system. Subsequently, the effectiveness of the proposed techniques is shown for the state estimation of the lithium-ion battery cell. With both techniques, the estimation accuracy can be significantly enhanced. Furthermore, they provide a framework for systematically designing TNL interval observers.
In language technology, clean data is fundamental for training high-quality models, yet large corpora often contain substantial noise due to OCR errors, missing diacritics, and various user-generated inconsistencies. This paper presents a comprehensive text cleaning pipeline tailored for Hungarian, leveraging transformer-based language models optimized for three key tasks: OCR error correction, diacritic restoration, and filtering grammatically incorrect sentences. We introduce huT5, a Hungarian adaptation of the mT5 model, which reduces model parameters and resource demands while maintaining strong performance on Hungarian-specific text cleaning tasks. The huT5 models were fine-tuned on carefully constructed Hungarian corpora for each task and benchmarked against state-of-the-art methods, demonstrating competitive results, particularly in OCR error correction and diacritic restoration. Our pipeline offers an efficient, freely accessible solution to enhance data quality for Hungarian NLP applications, setting a new standard in resource-efficient, language-specific text cleaning.
