This special issue of the International Journal of General Systems (IJGS) includes peer-reviewed extended versions of a selection of papers presented at the 11th Workshop on Uncertainty Processing (WUPES'18 held in Třeboň, Czech Republic, in June 2018).
The WUPES series of workshops began in 1988 and has been continued every three years since then at various locations in the Czech Republic. Throughout its 30-year history, the workshop has become a prestigious event that attracts scientists from around the world who are interested in artificial intelligence. As indicated in the title, the workshop has a broad scope covering all aspects of processing information under uncertainty. This topic is common to all areas of artificial intelligence. Thanks to the very friendly and informal atmosphere and a specific audience with a wide range of interests, the workshop offers a place to foster creative intellectual activities and the exchange of ideas. All accepted papers are presented in a plenary oral presentation – lecture – with a generous time frame available. A lot of time is devoted to very stimulating discussions.
In 2018, a total of 19 papers on machine learning, information theory, belief function theory, fuzzy logic, statistics, operations research, management science, and stochastic analysis were presented. In the selection process for this special issue, we tried to capture the rich variety of the presented methodological approaches and the quality of the selected papers was judged by the members of the Program Committee in accord with the usual practice of IJGS. There are, however, many other articles that would have also deserved to be included in this special issue.
Eventually, after a thorough reviewing process, the following six papers (extended versions of the conference papers) were included in this special issue:
Capotorti shows in “Probabilistic Inconsistency Correction for Misclassification in Statistical Matching, with an Example in Health Care” how the mixed integer programming helps in incoherence correction procedure to the misclassification paradigm of the statistical matching. The method is illustrated on a prototypical example from the health care analysis with information coming from two different sources.
“Detecting Correlation Between Extreme Probability Events” by Coletti, van der Gaag, Peturitti, and Vantaggi studies a new definition of correlation that does not lead in counter-intuitive results in the case of extreme probability events being present. This study allows us to handle extreme probability events using different levels of strength of the zero probability.
A new method for learning parameters of a probabilistic model from small datasets is presented by Plajner and Vomlel in “Learning Bipartite Bayesian Networks under Monotonicity Restrictions” and its properties (particularly concerning monotonicity) are analyzed. The method is compared with other standard methods and, in the case of small datasets, it outperforms them.
Shenoy investigates the so-called expected value in the case of belief functions and he introduces its new definition in “An Expectation Operator for Belief Functions in the Dempster-Shafer Theory”. Note that all the previous definitions assumed a transformation of the respective belief function into a probability function and then the standard definition of expected values was used. Transforming a belief function onto a probability function involves loss of information. This new operator works directly with belief functions.
“A Note on Approximation of Shenoy's Expectation Operator Using Probabilistic Transforms” by Jiroušk, Kratochvíl, and Rauh is an extensive empirical study of the above-mentioned Shenoy's expectation operator with respect to other indirect approaches, including various probabilistic transforms.
Alsuwat et al. (in “Adversarial Data Poisoning Attacks Against the PC Learning Algorithm”) deal with the importance of data integrity as a key component to Bayesian network structure learning algorithms. They show how an adversary could generate the desired network with the aid of a PC algorithm and they explore and analyze what is the minimum number of changes in the data that leads to a desired output of the PC algorithm.
To conclude these introductory words, we want to sincerely thank Radim Bělohlávek for his support in the many stages of preparing this issue. Our gratitude is also extended to the reviewers for their interesting comments and constructive remarks, which often helped the authors substantially improve their papers.