Classification Paradigms
Classification systems have the task of assigning an input sample to one of a set of classes, using a priori knowledge usually given in the form of a training set, i.e. a set of samples manually labeled with the correct class. The MIVIA lab research activities on classification paradigms have been focused on two aspects:
- Multiple Classifier Systems;
- Classifiers with a reject option.
In Multiple Classifier Systems, the idea is to improve the classification performance by combining several, possibly complementary, individual classifiers, so as to take advantage of the specific conditions to which each classifier is most suited, and to compensate the individual weaknesses.
A classifier with a reject option is a classifier that can defer a particular input sample, to a human operator, or to another classification system; this rejection has a cost, but can be convenient if this cost is less than the cost of an error, and the system is able to reject mostly samples that would have been misclassified.