Filipakis Panagiotis (Phd Candidate)

Thesis title: Data reduction for complex training data

Supervisor: Ougiaroglou Stefanos

Advisory Committee Members:

Dimitris A. Dervos , Professor, IHU
Geogios Evangelidis, Professor, University of Macedonia

Abstract:

The goal of the dissertation is the development of techniques for reducing the complex training data for instance-based classification. By the term “complex data”, we refer to data with multiple labels, data streams, data in non-Euclidean spaces, distributed data sets, etc. There are many data reduction techniques available in the literature for classification problems. These techniques either select prototypes (representative instances) (Prototype Selection) or generate prototypes by summarizing similar instances (Prototype Generation). The majority of these techniques address classification problems with relatively simple data, i.e., numerical data organized in rows and columns. In contrast, the research aims to the development of new data reduction techniques for complex data.