Rough Sets in Perception-Based Computing

Intelligent systems for many real life problems can be modeled by systems of complex objects and their parts changing and interacting over time. The objects are usually linked by certain dependencies, can cooperate between themselves and are able to perform complex and flexible actions (operations) in an autonomous manner. Such systems are identified as complex dynamical systems [2,40], autonomous multiagent systems [20,40], or swarm intelligent systems (see, e.g., [28,7]). One of the main challenges to be solved in intelligent systems research is the development of methods for approximate reasoning from measurements to perception, i.e., from concepts that can be directly approximated using sensor measurements to concepts, expressed by human beings in natural language, that are the perception results [42]. The existing methodologies and technologies are not adequate to solve many problems associated with this challenge. Among such problems are, e.g., classification and understanding of medical images [30], control of autonomous systems such as unmanned aerial vehicles or robots (see, e.g., [47,44]) or problems pertaining to monitoring or rescue tasks in multiagent systems [11]. Nowadays, new emerging computing paradigms are investigated in an attempt to develop methods for solving such problems. The further progress depends on a successful cooperation of specialists from different scientific disciplines such as mathematics, logic, philosophy, computer science, artificial intelligence, biology, physics, chemistry, bioinformatics, medicine, neuroscience, linguistics, psychology, and sociology. In particular, different aspects of reasoning from measurements to perception are investigated in psychology [1,3,16], neuroscience [30,24], theories of cognition [21], layered learning [39], mathematics of learning [30], machine learning, pattern recognition [14], data mining [17] and also by researchers working on recently emerged computing paradigms, like computing with words and perception [43], granular computing [25], rough sets, rough mereology, and rough-neural computing [25]. In this lecture we overview some of these new computing paradigms and some of the interactions between the various disciplines that have been mentioned. The concept approximation problem is the basic problem investigated in machine learning, pattern recognition [14] and data mining [17]. It is necessary to induce approximations of concepts (models of concepts) from available experimental data. The data models developed so far in such areas as statistical