Semi - supervised Learning Methods for Data Augmentation

The original goal of this project was to investigate the extent to which data augmentation schemes based on semi-supervised learning algorithms can improve classification accuracy in supervised learning problems. The objectives included determining the appropriate algorithms, customising them for the purposes of this project and providing their Matlab implementations. These algorithms were to be used to develop a robust system for achieving data augmentation in arbitrary application areas. For evaluation purposes, a general framework for assessing the quality of data augmentation achieved was to be constructed. The project met and exceeded all of the success criteria. A survey of theoretical results underlying data augmentation has been conducted. Full, general implementations of Bayesian Sets, Spy-EM and Roc-SVM algorithms, as well as their proposed extensions have been implemented. A general scheme for achieving data augmentation in binary and multi-class classification has been developed and successfully applied to the three application areas proposed. An evaluation framework for assessing the quality of data augmentation was implemented and used to give statistical significance to the results obtained. i Special Difficulties None. ii Declaration I, Nikola Mrksic of Trinity College, being a candidate for Part II of the Computer Science Tripos, hereby declare that this dissertation and the work described in it are my own work, unaided except as may be specified below, and that the disser-tation does not contain material that has already been used to any substantial extent for a comparable purpose. I give permission for my dissertation to be made available in the archive area of the Laboratory's website.