Effective Techniques of the Use of Data Augmentation in Classification Tasks
Unai Garay-Maestre · RUA, Repositorio Institucional de la Universidad de Alicante (Universidad de Alicante) · 2018
This report follows the research and development of a final degree project of computer engineering. The purpose of this project is to accomplish a new method to overcome the lack of data. In the literature the strategy that is accustomed to achieve this task is data augmentation which is a method that artificially creates new data based on the modifications of the existing data. The heuristics underlying this modifications are very dependent on which processes are suitable for the classification task at issue. In this project we introduce an alternative using Variational Autoencoders which are powerful generative models. These are capable of extracting latent values from input variables to generate new information without the user having to take specific decisions.