Decision fusion in cooperative adaptive systems

Summary form only given. Research on cooperative, adaptive intelligent systems, involves studying, developing and evaluating architectures and methods to solve complex problems using adaptive and cooperative systems. These systems may range from simple software modules (such as a clustering or a classification algorithm) to physical systems (such as autonomous robots, machines or sensors). The main characteristic of these systems is that they are adaptive and cooperative. By adaptive, it is meant that the systems have a learning ability that makes them adjust their behaviour or performance to cope with changing situations. The systems are willing to cooperate together to solve complex problems or to achieve common goals. In pattern recognition, there are notable contributions on the use of multiple classifiers. The most dominant decomposition model used is an ensemble of classifiers (identical structures) that are trained differently. Most of the innovations are in the combining methods. There are weighting and voting approaches, probabilistic approaches and approximate and fuzzy logic approaches. In the area of sensor fusion, there have been some interesting ideas for fusing the data and decisions of the sensors. However, most of these combining schemes are usually applied as a post processing stage. In this work are concerned with investigating architectures and methods of aggregating decisions in a multi-classifier or multi-agent environment. New architectures that allow active cooperation are developed. The classifiers (or agents) have to know some knowledge about others in the system. Different forms of cooperation are reported. In order for these architectures to allow for dynamic decision fusion, the aggregation procedures have to have the flexibility to adapt to changes in the input and output and adjust to improve on the final output. Changes are learned by means of extracting features using feature detectors. Applications of these architectures to problems in classification of data, distributed data mining and clustering are illustrated.