On Thursday April 7th 2011 at 15:30 in room 52.321, Zbigniew Ras, from the University of North Carolina and Warsaw University of Technology, will give a research seminar on "Cascade classifiers for automatic music indexing".

Abstract: In a hierarchical decision system S, a group of classifiers can be trained using objects in S partitioned by values of the decision attribute at its all granularity levels. Then, attribute values only at the highest granularity level (corresponding granules are the largest) are used to split S into decision sub-systems where each one is built by selecting objects in S of the same decision value. These sub-systems are used for training new classifiers at all granularity levels of its decision attribute. Each sub-system is split further by sub-values of its decision value. The obtained tree-type structure with groups of classifiers assigned to each of its nodes is called a cascade classifier. In the area of automatic music indexing, this cascade classifier makes a first estimate at the highest level of decision attribute values, which stands for the musical instrument family. Then, the further estimation is done within that specific family range. Experiments have shown better performance of a cascade system than traditional flat classification methods which directly estimate the instrument without higher level of family information analysis. Also, we will introduce the new hierarchical instrument schema according to the clustering results of acoustic features. This new schema better describes the similarity among different instruments or among different playing techniques of the same instrument. The classification results show the higher accuracy of a cascade system with the new schema compared to the traditional schemas.

Zbigniew Ras website