By Jim Bezdek, U. of Melbourne and Bernadette Bouchon-Meunier, U. Paris
By Boris Kovalerchuk, Central Washington University and Vladik Kreinovich, University of Texas
To “warm up” this round table, the WCCI 2012/FUZZ-IEEE 2012 program has scheduled a tutorial “Fuzzy Logic, Probability, and Measurement: Similarities and Differences in Computing with Words” organized by B. Kovalerchuk. This free tutorial will pave the way for the round table. All WCCI participants are invited to this tutorial. The tutorial will help to shorten the introductory part of the round table, and to concentrate on a deeper discussion with more participants. This round table is envisioned as an extension of the round table that B. Bouchon-Meunier organized on uncertainty modeling at the World Conference on Soft Computing in 2011 with invited panelists (Zadeh, Widrow, Kacprzyk, Kovalerchuk, Perlovsky) with multiple requests from the audience to extend it. One of the specific flavors of the panel will be the analysis of different approaches from the multiple perspectives keeping a neutral “denominational” stance. One of such perspectives is the scientific rigor. The optimistic view is that high scientific rigor is compatible with the high uncertainty of CWW. Other perspectives are the time horizon to find the solution of the task, and the importance of the task. Do we have a few minutes or a few years to generate a solution? These constraints impact very significantly the desired and achievable rigor of the solution.
The panel will cover the broad range of performance metrics that go beyond the simple calculation of success rate or accuracy on a limited test set. These include sensitivity, specificity, positive and negative predictive value, receiver operating characteristic (ROC) curve, and area under the ROC curve (AUC). Particular attention will be paid to the ROC curve formalism and its extension beyond the binary categorization problem. Both parametric and nonparametric uncertainty calculation techniques will be discussed, with particular emphasis on jackknife and bootstrap methods. The importance of good data hygiene in maintaining complete separation of training and test data will be discussed, and the significance of understanding training as well as test-set variability (and human observer variability where appropriate) will be examined.
All of these methodologies have been available for some time, and yet many papers at this and other conferences pay scant attention to any kind of rigorous evaluation of the results they report. The panel will consider how to encourage the better use of these methods in the CI community.
Over the last two decades there has been a growing interest in the need for designing intelligent systems to address complex control and decision making problems. One of the most challenging issues for the intelligent system is to effectively handle real-world uncertainties that cannot be eliminated. The control-related uncertainties include sensor imprecision, instrumentation and process noise and disturbances, unpredictable environmental factors; while the decision-making-related uncertainties include risk analysis, human factors, safety and security evaluation, to name a few. These uncertainties result in a lack of the full and precise knowledge of the system including its state, dynamics, and interaction with the environment in control applications and a lack of transparency and better understanding of decision analysis in society and policy support. Computationally intelligent systems, based on different techniques, including fuzzy logic, neural networks, genetic algorithms and others, have shown great potential to solve these demanding, real-world problems that exist in uncertain and unpredictable environments.
To be ready for new future applications, we must now consider:
Panelists will consider these and other questions, offering to the audience points of view based on their own experience.
Preliminary list of panelists
Robert Kozma and Jennie Si (Chairs)
Haibo He, Janusz Kaczprzyk, Jim Keller, Luis Magdalena, Marios Polycarpou, Lipo Wang
Computational Intelligence is a relatively new research field. A lot of educational materials have been created in various fields of CI in the past decades. However, due to the field's relatively youth, its fundamental achievements has not been organized into a comprehensive curriculum yet. It is crucial for the development of the field to have high-quality educational materials on the state of art of CI. This allows attracting and educating talented and enthusiastic students and documenting the progress in the field. The panel will discuss various areas of CI education, including existing databases and course materials, online resources and video lectures, development of new textbooks, open-source software, and others. Various recommendations for future actions will be discussed as well.