Empirical Evaluation of Variational Autoencoders for Data Augmentation

• DOI: 10.5220/0006618600960104
• Publication Date: 2018-02-26T14:21:06Z
• Authors: Javier Jorge; Jesús Vieco; Roberto Paredes; Joan Andreu Sanchez; José Miguel Benedí
• Keywords: Variational Autoencoder; 0202 electrical engineering, electronic engineering, information engineering; Generative Models; 02 engineering and technology; Data Augmentation; LENGUAJES Y SISTEMAS INFORMATICOS

Since the beginning of Neural Networks, different mechanisms have been required to provide a sufficient number examples avoid overfitting.Data augmentation, most common one, is focused on generation new instances performing distortions in real samples.Usually, these transformations are problem-dependent, and they result synthetic set of, likely, unseen examples.In this work, we studied generative model, based paradigm encoder-decoder, that works directly data space, is, with images.This model encodes input latent space where will be applied.After completing this, can reconstruct vectors get samples.We analysed various procedures according could carry out, as well effectiveness process improve accuracy classification systems.To do use both original after reconstructing altered version vectors.Our results shown using pipeline (encoding-altering-decoding) helps generalisation classifiers selected.