Computer Engineering Department, Biomedical Engineering Institute
Chiang Mai University, Chiang Mai 50200, Thailand

Abstract:

Data Analysis is a process to analyze data in terms of representing, describing, evaluating, interpreting the data using statistical methods. Data can come in the form of statistical representation or a vector of numbers in which numeric pattern recognition algorithms can deal with this type of data set. Another type of data can be in the form of syntactic data. For this type of data set, there is another research branch in pattern recognition called syntactic pattern recognition that is able to analyze it. Each sample in syntactic data set is normally represented as a string. The strings in the same data set can have different lengths. Also, the string does not have any mathematical meaning that we can calculated as if they are vectors of numbers. One of the popular theories used in data analysis is Fuzzy set theory, an extension of the classical set introduced by Lotfi Zadeh in 1965. Since then, there are many theories and applications developed based on Fuzzy set theory. In this talk, the utilization of the Fuzzy Set in data analysis where there are uncertainties in the data set will be summarized. All algorithms in this talk are developed at Computational Intelligence Research Laboratory, Chiang Mai University. We will show applications of these algorithms in several real-world problems.

KERMIT, Department of Data Analysis and Mathematical Modelling
Ghent University, Coupure links 653, 9000 Gent, Belgium

Abstract:

The theory of fuzzy sets, nearing its sixtieth anniversary, faces a challenging landscape marked by a proliferation of generalizations lacking semantic clarity and robust elicitation procedures. Similarly, numerous variants of fuzzy decision-making methods populate the literature, often characterized by ad hoc choices and unproven claims of enhanced decision-making capabilities. These developments not only fail to advance the field but also risk tarnishing its reputation in domains such as machine learning and operations research. Amidst these challenges, it is crucial to revisit foundational contributions that have stood the test of time. The compositional rule of inference and the extension principle introduced by Lotfi Zadeh in the seventies offer enduring insights, albeit often overlooked in contemporary discourse. Additionally, Goguen’s early recognition of lattice theory as the appropriate framework for fuzzy set theory remains as relevant today as ever. Drawing from these foundational principles, this talk explores pivotal milestones in fuzzy set theory, including fuzzy relational equations, computation with fuzzy quantities, and convolution lattices. These concepts not only address the propagation of non-stochastic uncertainty but also provide a lens through which to critically assess recent developments in the field. By reexamining these fundamental concepts and their applications, we can navigate the complexities of modern fuzzy set theory with renewed clarity and purpose. This journey not only enriches our understanding of uncertainty modeling but also opens new avenues for innovation in fields reliant on fuzzy logic.

Sorbonne Université - LIP6
75252 Paris CEDEX 05, France

Abstract:

TBD

Laboratory for Autonomous and Mobile Systems
University of Ljubljana, SI-1000 Ljubljana, Slovenia

Abstract:

Traditional approaches in computational intelligence and machine learning often rely on two key assumptions: the availability of a sufficiently large historical dataset for model training, and that the operational environment will present data instances similar to those seen during training. However, these assumptions frequently break down in real-world scenarios, particularly in autonomous mobile robotics, where data may be scarce, environments are dynamic, and system behavior can change over time. Additionally, as the volume of incoming data grows, conventional iterative algorithms—which depend on multiple passes over the data—become impractical and computationally expensive. Evolving systems, designed for online and incremental learning from data streams, offer a promising solution. These systems adapt in real time by performing single-pass learning, evolving and pruning model components on demand, and accommodating non-stationary behavior without retraining from scratch. This presentation explores the application of evolving learning frameworks in the context of different application areas for real-time identification. monitoring, decision-making, and control. Emphasis will be placed on practical implementations, highlighting how these models can incrementally adapt to complex, changing environments while maintaining computational efficiency and robust performance.