Pozvané Prednášky

Organizational Knowledge and its Managerial Applications

Jan Voráček

Faculty of Management, University of Economics
Jarošovská 1117/II, 377 01 Jindřichův Hradec, Czech Republic
e-mail: voracekj@fm.vse.cz

Engineering knowledge is usually considered as a set of sentences, generated from a certain formal language in a sound way like graphs or rules. The main advantages of this viewpoint are (i) closeness of the processed domain and (ii) straightforward validation of resultant structures. On the other hand, the preselected way of knowledge representation logically narrows its overall expressiveness and biases user understandability.

Institutions, however, consider knowledge as tool for overall performance increasing. As a typical enterprise is a complex socio-technical system, it is evident that also the related knowledge must reflect the human dimension and team character of internal processes. Consequently, the tools and method used across the organizational knowledge engineering lifecycle differ from those handling just a narrow niche of it. Beyond the natural user friendliness, the related structures and languages must be able to handle theoretically problematic phenomena of nonstationarity, subjectivity or uncertainty.

The talk describes how the tacit, distributed and continuously evolving organizational knowledge can be elicited, represented, processed and validated. Theoretical considerations are illustrated with simulation models of typical healthcare management cases.

Keywords: organizational knowledge, knowledge engineering lifecycle, simulation model

Set Approximations in Qualitative Data Analysis

Milan Vlach

Faculty of Mathematics and Physics, Charles University
Malostranské náměstí 25, 118 00 Praha 1, Czech Republic

Kyoto College of Graduate Studies for Informatics, 7, Monzen-cho, Tanaka,
Sakyo-ku, Kyoto, 606-8225 Japan

e-mail: milan.vlach@m .cuni.cz

When modeling complex phenomena, various simpli cations and idealizations are indispensable, and when analyzing the resulting abstract models, fur- ther approximations of their mathematical components are frequently needed. Because in many such instances it is impossible to describe sets precisely, a substantial part of qualitative data analysis is concerned with set approximations. The mathematical background of set approximations used in qualitative data analysis is based on the interplay among relational structures and set-valued set functions. A number of known classical set functions have appeared in the literature of rough set theory and its applications. The main aim of the talk is to show in a clear and lucid way how some well established algebraic, logical, and topological structures are closely related or identical to those introduced, discovered and rediscovered in the eld of rough sets.

Keywords: Qualitative data analysis, Formal concept analysis, Approximation operators, Rough sets

Practical Feature Selection in Statistical Pattern Recognition

Pavel Pudil (1), Petr Somol (2)

(1) Faculty of Management, Prague University of Economics, Czech Republic
e-mail: pudil@fm.vse.cz

(2) Institute of Information Theory and Automation, Czech Academy of Sciences
e-mail: somol@utia.cas.cz

A broad class of decision-making problems can be solved by learning approach. This can be a feasible alternative when neither an analytical solution exists nor the mathematical model can be constructed. In these cases the required knowledge can be gained from the past data which form the so-called learning or training set. Then the formal apparatus of statistical pattern recognition can be used to learn the decision-making. The rst and essential step of statistical pattern recognition is to solve the problem of feature selection - or more generally - dimensionality reduction.

The methodology of feature selection in statistical pattern recognition will be overviewed with particular emphasis put on the practitioner's view. Therefore, among the broad variety of available tools we will focus only on those tools that have been proven most usable in real applications.

The main aspects of the Feature Selection problem, with direct connection to classi cation tasks, will be covered. Most recently discussed topics and related problems will be addressed as well, including the problem of feature over-selection, feature selection stability, and automated determination of the suitable feature subset size.

A new open-source software library specialized on the feature selection problem will be presented. The Feature Selection Toolbox 3 library architecture, included tools, usage scenarios and use cases will be presented. The accompanying web portal will be introduced as a practical resource available to researchers and practitioners in wide variety of elds.

Keywords: pattern recognition, feature selection, dimensionality reduction, software library, machine learning.