Marek Gagolewski
Deakin University, School of Information Technology, Melbourne, Australia
Title: Clustering and Aggregation
We will examine a few scenarios where aggregation methods can aid in the
construction, analysis, and evaluation of tools related to data
clustering, including linkage criteria, partition similarity measures, and
cluster validity indices. We indicate some noteworthy challenges for both
theoretical and practical future research endeavours.
Libor Běhounek
University of Ostrava, CE IT4Innovations - IRAFM, Ostrava, Czech Republic
Title: Making the domains of fuzzy sets explicit
The usual definitions of fuzzy set operations assume that the involved
membership functions have the same domain of definition. The common way
of dealing with fuzzy sets defined on different domains is to fill in
zeros (or other suitable values) outside the domains of definition. It
can be shown, however, that such modification may introduce spurious
values into the results unless the domains of membership functions are
explicitly taken into account. As a remedy, I will compare several
methods of explicitly marking the domains of membership functions and
discuss several ways of extending fuzzy set operations to accommodate
the explicit domains. I will present selected results on such
variable-domain operations on fuzzy sets and fuzzy relations and hint at
a few applications in function approximation and fuzzy rules. (Most of
the presented results come from joint work with Martina Daňková.)
Martin Holeňa
The Czech Academy of Sciences, Institute of Computer Science, Prague, Czech Republic
Title: Machine Learning Alleviates the Dilemma of Black-Box Optimization
Nowadays, real world optimization problems more and more frequently optimize black-box objective functions, which are not evaluated analytically, neither explicitly nor implicitly, but rather empirically by simulations, measurements or experiments. Most successful in their optimization are stochastic optimization methods, especially evolution strategies, as they make nearly no assumptions about the black-box objective, tend to find its global optimum, and need only its empirically obtained values, in as many points as possible. On the other hand, the empirical evaluation of a black-box objective is often costly and/or time consuming, which makes desirable to evaluate it in as few points as possible. To alleviate that dilemma, a machine-learning-based approach has been used for nearly two decades: a model is learned using data from previous iterations of the optimization method and serves as a surrogate of the true black-box objective functions in most of its new evaluations. The talk will give a survey of successful surrogate models, as well as a survey of strategies how to control when to evaluate the true objective and when its surrogate.
Manuel Ojeda-Aciego
University of Malaga, Malaga, Spain
Title: Galois connections between fuzzy unbalanced structures
Galois connections are pervasive structures which have found application in several areas. In our case, we found them in an important construction in (Fuzzy) Formal Concept Analysis.
We focus on the problem of characterizing the existence of a fuzzy (isotone/antitone) Galois connection under certain circumstances. Specifically, given a set with a certain structure (fuzzy preposet/poset/T-digraph) and a function from it to an unstructured set, we look for necessary and sufficient conditions to give structure to the codomain and find the adjoint of the function.
The path followed aims at considering a properly fuzzy Galois connection in its most general sense. This leads to a twofold objective: on the one hand, the components of the connection should be fuzzy functions (in some approaches, although the structure of the domain is fuzzy, the pair constituting the connection consists of crisp functions); on the other hand, the underlying fuzzy structure requires certain minimal conditions so that the construction do not collapse.
We will present a survey of results obtained in different instances of the problem above.
Lenka Halčinová
Pavol Jozef Šafárik University in Košice, Košie, Slovakia
Title: The concept of conditional aggregation
The theory of aggregations is nowadays an important part of mathematics. In the contribution, we will introduce the concept of conditional aggregation operators. We define novel survival functions and the generalized Choquet integral based on them. We will discuss their properties and indicate their possible advantages in applications.