Computational Intelligence in Decision-like Procedures
(Slovak National Agency for Science - under registration # 9433 - 1999-2001)
Computational Intelligence represents a part of Artificial Intelligence and mainly integrating 3 different technologies concerning artificial neural networks, fuzzy systems and evolutionary systems. Integration of these systems results in so called hybrid intelligent systems. The project covers and focuses on basic-and-fundamental research issues and also application domain in the following directions: intelligent control, intelligent prediction systems and intelligent image processing systems. In all application problems we focus on decision-like procedures and utilization of obtained knowledge base in these processes. The level of decision procedures is the crucial point of the technology level. If we will be able to make intelligent decisions (decisions with high level reliability) we can improve many technological processes in general. The machine IQ seems to be an attempt to measure the level of intelligence of human-made systems and to evaluate decision procedures in technology.
In the domain of intelligent control we focus the problems of intelligent mobile robot control with aim to solve tasks as navigation in unknown environment, obstacle avoidance and progress to highly complex problems including robot soccer. The project intent to use Lego-robots (Lego-mindstorm) and run number of verification projects concerning fuzzy control, neurocontrol, GA-based control and hybrid systems. ( In this part of the project the following colleagues are involved : Jan Vascak, Rudolf Jaksa, Miroslav Hudec, Peter Kostelnik, Marek Samulka, Gyongy Gebeova.)
In the domain of prediction systems we consider prediction as so called trends classification/prediction/decision related to time series. In fact we are dealing with the problem of function extrapolation. Usually prediction of selected variable depends on various inputs. General we can divide analysis concerning prediction procedure into 2 parts : Technical Analysis and Fundamental Analysis. The prediction can be influenced by factors from technical as well as fundamental analysis. The prediction of processes where factors from technical analysis play the lye role is the main focus of this project. Predictions of processes where factor from fundamental analysis are important is very difficult and are not considered in this project. The key issue of the project is to study FIR neural networks with aim to avoid determination size of input window of time series data. We use data from Santa-Fe international database and also some data from power engineering domain with goal to predict the electricity load. Electricity load forecast is very important part of decision system to save expenses in electricity production domain. ( In this part of the project the following colleagues are involved: Stanislav Kaleta, Daniel Novotny, Martin Kropuch).
In the domain of image processing is classification procedure will be focused with improving the quality of decision about pixel membership to the class of interests. Classification and object identification procedures are tested based on decision procedures are being tested, analyzed and development of more sophisticated decision systems is underway mainly based on computational intelligence tools. From neural classifiers the main focus was/is on ARTMAP-like systems and development of MF-ARTMAP systems that represents adaptive/learning system for decision procedure with identification of membership function of fuzzy clusters/classes in multidimensional feature space. (In this part of the project the following colleagues are involved: Marek Samulka, Norbert Kopco, Richard Valo).