The 12th International Conference on Fuzzy Systems and Data Mining (FSDM 2026)

Abstract: As a generalization of the classical Lebesgue integral, set-valued integrals had been greatly developed, and widely applied in the fields such as Mathematical Economy, Control Theory, Optimization Theory. In recent years, based on nonadditive integrals, various kinds of set-valued nonadditive integrals were developed. As a powerful tool for modeling nondeterministic problems and a typical nonadditive integral, the Choquet integral has attracted much attention. This lecture gives an overview of several existing set-valued Choquet integrals: (i) Choquet integrals of set-valued functions with respect to set-valued fuzzy measures, which include the special cases of Choquet integrals of set-valued functions with respect to fuzzy measures and of single-valued functions with respect to set-valued fuzzy measures; (ii) Generalized Choquet type set-valued integral (GCTSI) of single-valued functions with respect to set-multifunctions and additive measures; (iii) Double set-function multi-valued Choquet integral (DSMVCI) for single-valued functions with respect to set-multifunctions and fuzzy measures. Various kinds of their properties and convergence theorems are shown, and some set-valued type inequalities, such as Jensen's and Markov's are given. Related to many applications in different fields, such as decision making are provided.

Abstract: An Ω‑algebra is a kind of fuzzy algebra in the lattice-valued framework, with an algebra equipped with a lattice‑valued equality. Algebra identities are satisfied as special lattice-valued formulas, and as a consequence, cuts of the equality correspond to weak congruences of the underlying crisp algebra. In earlier work, we investigated many types of Ω‑algebras. Recently, in an attempt to study consequences of our work in the classical setting with the [0,1] interval, we have studied connections with chains in weak congruence lattices. Building on this foundation, we now turn to the classical setting of Ω‑groups and Ω‑quasigroups, where subgroup lattices and weak congruence lattices interact directly with the solvability of equations.
Classical group theory provides deep results on solvability: equations in groups are closely tied to subgroup chains, normal series, and congruence relations. In particular, solvability of equations often reflects the structure of chains in subgroup lattices, such as composition series or chains of normal subgroups. These chains serve as algebraic witnesses to the existence of solutions, linking lattice‑theoretic properties with equation solvability. Similarly, in quasigroups, congruence chains impose constraints on the solvability of functional equations, revealing structural parallels with group theory.
Our contribution demonstrates that chains of weak congruences in Ω‑groups can be systematically aligned with chains in subgroup lattices, thereby providing a lattice‑valued framework for analyzing solvability of equations. Extending this approach to Ω‑quasigroups, we show how weak congruence chains capture algebraic conditions that govern equation solvability in non‑associative settings. These results enrich the representation theory of Ω‑algebras by connecting chains of weak congruences with classical solvability theory, offering a unified perspective that bridges discrete subgroup lattices, weak congruence structures, and equation systems in both groups and quasigroups. In this way, by dealing with numbers, we are able to approximately solve some types of equations.

Keywords: Ω‑algebra, fuzzy algebra, approximate solving of equations

Abstract:
Volterra integral equations (VIEs) are important mathematical tools for modeling systems with memory and have applications in various fields of science, engineering, and applied mathematics. This talk begins with classical (crisp) VIEs and their numerical approximation, before extending the discussion to fuzzy Volterra integral equations, which allow uncertainty in the mathematical description of real-world systems. We then consider a more general class of fuzzy VIEs involving proportional delay and piecewise discontinuous kernels. In this setting, the discontinuity curves are defined by the proportional delay parameter. This structure partitions the original integral equation into several subintervals, each associated with its own continuous kernel and nonlinear term. We also establish the existence and uniqueness of solutions using a Picard-type iteration and the Gronwall inequality. For the numerical solution, we develop a collocation method based on piecewise fuzzy polynomial approximation and apply the Gauss--Legendre quadrature rule separately on each subinterval. We also derive an error bound in terms of the fuzzy modulus of continuity and investigate the performance of the proposed method through numerical examples. The results illustrate the accuracy and effectiveness of the approach for solving fuzzy Volterra integral equations with proportional delay and piecewise discontinuous kernels. The talk thus provides a systematic progression from classical Volterra integral equations to fuzzy formulations and, finally, to delay problems with discontinuous kernels, emphasizing the theoretical and computational challenges at each stage.

Abstract: The rapid development of generative artificial intelligence (GenAI), extended reality (XR), and quantum computing (QC) creates new opportunities for adaptive and immersive learning, but technological combination alone does not guarantee pedagogical or computational value. The aim of the invited speech is to propose the Quantum-Generative Immersive Education (QGIE) conceptual architecture, based on the principle of functionally justified technological integration. The architecture organizes learning as a closed adaptive loop linking learner interaction, a dynamic learner model, pedagogical decision-making, computational requirement extraction, technology-resource selection, GenAI adaptation, and XR realization. A Quantum Justification Framework (QJF) is proposed, introducing a mandatory classical baseline, feasibility and quantum relevance screening, multicriteria comparison, and a Quantum Justification Index (QJI). The principal contribution is the shift from technology-driven integration toward a pedagogy-first, function-first, and technology-selective architecture in which a quantum resource is activated only when it provides sufficiently justified functional value relative to the applicable classical alternative.

Abstract: Accurate traffic forecasting requires effective modeling of temporal dynamics, spatial interactions, and heterogeneous data sources. This invited talk presents our recent studies on multimodal fusion and efficient spatiotemporal modeling for traffic forecasting. We first investigate multimodal metro flow prediction under distributed and privacy-sensitive settings, where cross-modal alignment, efficient long-range sequence modeling, and collaborative learning are jointly considered to improve predictive robustness. We then explore adaptive traffic flow forecasting by combining multi-scale temporal representations, multi-granularity spatial graph construction, and cross-domain feature interaction to better capture complex spatiotemporal dependencies. Moving toward finer-grained traffic analysis, we further study lane-level forecasting with dynamic spatial memory and bidirectional knowledge transfer, aiming to improve both modeling efficiency and adaptability across traffic scenarios. Finally, we broaden the discussion to vision-spatiotemporal fusion, examining how visual observations can be aligned with structured traffic dynamics at different representation levels, and highlighting the semantic and geometric gaps that remain in current forecasting systems. Together, these studies illustrate a progression from multimodal representation and adaptive spatiotemporal learning toward more robust, generalizable, and spatially aware traffic forecasting.

Abstract: Enhancing the resilience of electricity transmission infrastructure requires timely situational awareness, reliable diagnostics, and actionable early warnings under uncertain and heterogeneous conditions. The presented results are achieved from an integrated research effort focused on digital technologies and analytical models for critical infrastructure monitoring, with emphasis on experimental validation in controlled and pre-operational settings. It covers a modular approach combining remote sensing, field telemetry, operational logs, and physical simulation to support future digital twin and early-warning capabilities.
The presented work is derived from multiple data modalities and demonstrators. Airborne hyperspectral campaigns (VNIR/SWIR) over selected Slovak localities produced a georeferenced and annotated dataset enabling ecological and infrastructure mapping, including a labelled geodatabase of vegetation and non-vegetation structures. Complementary LiDAR point clouds were processed into classified outputs and canopy height models suitable for 3D integration. For condition monitoring, acoustic diagnostics assessed the feasibility of detecting patterns related to high-voltage insulators using large-scale audio recordings and power lines audio embeddings, while telemetric meteorological observations were analysed to identify conditions linked to moisture-driven degradation, and dust contamination. On the operational side, SCADA/RIS logs and PMU phasor measurements were investigated for synchronization, causality inference, and precursor detection using deep learning concepts (e.g., sequence models and graph-based architectures). Finally, fire-risk assessment was explored through Fire Dynamics Simulator scenarios to study vegetation heterogeneity effects and establish methodological foundations for calibration to local conditions.

Abstract: The digital transformation of decentralized healthcare demands integrated multi-omics datasets, population health records and translational evidence to support primary care clinical decisions. However, fragmented biological databases and disconnected analytical pipelines create major barriers to translating aging research into real-world healthcare practice. To tackle this challenge, we built AI-Aging, a unified AI-based digital infrastructure delivering an end-to-end translational pipeline spanning multi-omics profiling to population-wide aging health management. The system consists of four standardized, interoperable sub-platforms: HALDxAI, a large-scale aging knowledge graph containing 2.8 million biological nodes and 180 million relational links to dissect aging mechanisms; ARIES, a literature-derived evidence graph enabling traceable causal connections between geroprotective chemicals, molecular pathways and aging phenotypes; AITED, a translational database curating more than 8,700 aging interventions to measure translational gaps and safety signals for clinical trial design; and Fluxera, an interpretable predictive model trained on a cohort of 178,000 participants to map population aging trajectories and forecast individual health risks. Uniform data schemas, open APIs and traceable evidence chains are implemented across all modules, forming a closed-loop system for omics fusion, mechanistic interpretation, intervention assessment and population health forecasting. This infrastructure connects fundamental multi-omics findings with on-the-ground decentralized primary care, establishing a solid digital framework for precision aging intervention and equalized longevity health management.

Keywords: digital health transformation; multi-omics integration; AI knowledge graph; longitudinal health trajectory; ageing translational database; decentralized primary care; longevity research

Abstract: The integration of artificial intelligence, advanced sensing, and embodied intelligent systems is opening new opportunities to transform agricultural production and post-harvest management. This speech explores how AI-enabled sensing and machine learning can support early detection, prediction, and management of quality deterioration in stored pome fruits, with particular emphasis on the analysis of chemical volatiles and other non-invasive sensing signals.
By combining intelligent sensing with data-driven models, it becomes possible to move from conventional, reactive approaches to predictive and intelligent post-harvest management. AI systems can identify early indicators of deterioration, classify fruit conditions, optimise storage environments, and support decisions on when produce should be released to the market. These capabilities have direct relevance to embodied intelligence, where sensing, perception, decision-making, and physical action must operate together in complex real-world environments.
The speech will examine the broader industrial, economic, and societal implications of these technologies. For industry, intelligent early-warning systems can help protect large quantities of high-value agricultural produce and improve supply-chain efficiency. Economically, improved prediction and intervention can reduce avoidable post-harvest losses and associated financial costs. From a societal perspective, intelligent assessment of fruit quality can contribute to food security, sustainability, waste reduction, and safer decisions regarding food entering the consumer market.
The speech will conclude by considering how the convergence of AI, sensor fusion, and agricultural science can contribute to a new generation of intelligent, sustainable, and resilient food systems.