Cooperative Multi-Classifiers协同多分类器.pptVIP

Cooperative Multi-Classifiers By: Rozita Dara Supervisor: Prof. Kamel Outline Cooperative Systems Multiple Classifier Systems Cooperation in Multiple Classifier Systems Research Objectives Proposed Approach Experiments Results Future Work Cooperative Systems Cooperation: some kind of activity conducted between two or more entity, to common gain. “Cooperative work” was first introduced by Marx. It requires communication and awareness of others actions and outputs. System: an entity that maps an input to an output. Cooperative Systems: Systems are considered cooperative when they present team-based approach [Sommerville et al (1993)] Multiple Classifier Systems Multiple classifier systems consist of a set of classifiers and a decision combination function. Motivations: existence of many alternative solutions to a pattern recognition problem different patterns of misclassification by classifiers for the same data difficulty to find a single best classifier chance of loss of information Cooperation in MCS MCS is a system in which converts a set of input patterns into a set of predictions. Components of MCS cooperate to achieve a common goal, higher classification accuracy. Principles of designing cooperative MCS: communication and awareness of others actions. establishment of shared understanding and shared goals between the components. Cooperation in MCS (cont’d) Problem Statement: What type of cooperation among the components are most effective? Proposed Approach: A thorough understanding of cooperation among MCS components provides guidelines for optimization of the system. Cooperation in MCS (cont’d) Evaluation of Cooperation: identifying the components functionality of each component examining the effect of changes in components with respect to system performance identifying the type of interactions among components investigating various aspects of these interactions Research Objectives The focus of this research is on the sharing as